About 600 tools, ideas and settings from your 105 videos, sorted into 14 colour-coded chapters. You do not need to read all of it.
App, Skill, Repo (code on GitHub), Idea, Feature (built into a tool), Plugin, MCP (connects a tool to an agent), CLI (typed in a terminal), Model (an AI brain).
OpenAI's image maker, also usable through a Hermes plugin and a Codex skill.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
Makes images and video with major models through an MCP connector and command line, including face-consistent Soul.
Suits: Mantis. Mantis makes images and could generate them directly.
In your videos: Tristen O'Brien adds it to Claude as a custom connector · Named in 6 of your videos
Watch it here:
and 2 more videos
Free Adobe-style apps built in Rust, with a built-in MCP so agents can run them.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Mostly early alpha
In your videos: Named in 4 of your videos
Watch it here:
Turns plain English into editable HTML diagrams of how a system or workflow fits together.
Suits: Cyborg, Mantis. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches; Mantis could also use it.
In your videos: AI Edge cites 64.4K stars · Named in 2 of your videos
Watch it here:
A design editor, also available as an MCP in Hermes.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Canva's image model, with a Dream Lab tab and an API.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Makes social media graphics and saves them as PNG or PDF.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Claude's design canvas for mocking up apps and screens with colour options.
Suits: Mantis. Mantis does visual work, and Chris wants drafts made in Claude Design.
In your videos: Named in 2 of your videos
Watch it here:
Turns a prompt into an editable animation that you tweak with comments.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Beta for Team and Enterprise plans only
In your videos: Named in 2 of your videos
Watch it here:
Turns one sentence into a designed diagram, in about 45 types.
Suits: Mantis, Cyborg. Mantis does image prompts and creative work, which this produces; Cyborg could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Fast local image generation on Apple silicon Macs and iPhone.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Black Forest Labs image models for making pictures from text.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Watch it here:
xAI's image maker, available through an API and used for thumbnails.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Makes editable SVG images, which are sharp at any size.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
An image model that lets you edit inside the picture. In public beta.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Two image makers that turn text into pictures.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Gives agents Mobbin's library of app screens and flows for design ideas.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Google's image model, used to make a daily post image.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
An open-source shared design canvas with an agent. Bring your own key.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
A fast open image model with a Hugging Face Space to try it.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
A small 2 GB image model you can run on your own computer.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Free website that makes random manga-style maps and cities in your browser.
In your videos: Named in 1 of your videos
Watch it here:
The agent writes an HTML page that turns into an MP4 video.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Needs Node 22
In your videos: Named in 7 of your videos
Popularity: 71.8k stars
Watch it here:
and 3 more videos
Open-source tool that makes video from React code, with a map-animation skill.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Tristen O'Brien calls it the free-tier alternative to Claude Motion · Named in 7 of your videos
Popularity: 67.4k stars
Watch it here:
and 3 more videos
Cloud voices, music and lip-sync (Chimera), available through an API.
Suits: Alfred, Mantis. Alfred, the general assistant, could use it in everyday chats; Mantis could also use it.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Lets an agent control Blender to make 3D models and games.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Popularity: 32.3k stars
Watch it here:
Turns a project into a 20-second coded launch video; free on GitHub.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Lets an agent watch a video from a link or file.
Suits: Cerebra, Mantis. Cerebra does web and source-graded research and vault indexing, which this feeds; Mantis could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Lets Hermes control DaVinci Resolve Studio live for video editing.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Text-to-speech voices, including one steered by bracketed directions.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Animates 3D characters with loose clothing or animal bodies.
In your videos: Named in 1 of your videos
Watch it here:
A voice agent model with a free sandbox to try it.
In your videos: Named in 1 of your videos
Watch it here:
AI avatar video, combined with Remotion.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Copies the format of a viral video.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Nuno Tavares says skip: needs paid services
In your videos: Named in 4 of your videos
Watch it here:
Local text-to-speech voice. Slower than Piper.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
A game engine plus video model from Skywork, with an open Apache 2.0 licence.
In your videos: Named in 1 of your videos
Watch it here:
A video generation model that needs 8xH100 chips. The name may be wrong.
A voice agent model that handles several speakers and can call tools.
In your videos: Named in 1 of your videos
Watch it here:
A game engine sends a colour map and a video model draws it live.
In your videos: Named in 1 of your videos
Watch it here:
OBS streaming with WebSocket control, fed by the Moblin iPhone camera over SRT.
In your videos: Named in 1 of your videos
Watch it here:
An interactive world model you can move around in.
In your videos: Named in 1 of your videos
Watch it here:
Turns one image plus audio into a talking video, up to 5 minutes.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Free voice model covering 600+ languages, used behind Voice Studio.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Open-source video production with 12 pipelines and over 700 skill files.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Watch it here:
Turns product photos into videos inside ChatGPT.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
A tiny, fast text-to-speech voice that runs on your own machine.
Suits: Alfred. Alfred could voice replies cheaply, related to Chris's voice plans.
In your videos: Komputer Mechanic found it about 2x faster than Kokoro · Named in 2 of your videos
Watch it here:
ByteDance's model for making video from prompts.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Watch it here:
Models that make music from a text prompt.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Turns photos taken from several angles into a detailed 3D model.
In your videos: Named in 1 of your videos
Watch it here:
Three models that make video from prompts.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
A text-to-speech model. The caption was garbled.
In your videos: Named in 1 of your videos
Watch it here:
Free local voice cloning, scripts, dubbing and transcription. An alternative to ElevenLabs.
Suits: Mantis, Alfred. Mantis does image prompts and creative work, which this produces; Alfred could also use it.
In your videos: Claimed 61,000 stars · Named in 4 of your videos
Watch it here:
A mid-tier open video model. Slow on a Mac Studio.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
An open model that turns speech into text, so you can talk to your agents.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Tencent's world model that remembers where the camera has been.
In your videos: Named in 1 of your videos
Watch it here:
Anthropic's coding agent for the terminal, with skills, plugins, hooks, MCP and helper agents.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Pav Rusovs chose it over Hermes, citing Anthropic rules on subscriptions in third-party harnesses · Named in 39 of your videos
Popularity: 153.3k stars
Watch it here:
and 35 more videos
OpenAI's coding agent, in the terminal or the ChatGPT desktop app.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Samin Yasar uses Codex as the doctor that fixes Hermes · Named in 35 of your videos
Popularity: 128.1k stars
Watch it here:
and 31 more videos
An AI code editor with cloud agents. It can also use MCP tools.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
A library of interface parts that Claude can pull from.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
About 24 senior-engineer skills with commands to spec, plan, build, test, review and ship.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Popularity: 110.0k stars
Watch it here:
Open-source code review, with a VS Code extension.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Official skills for building Android apps with coding agents.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Google's code editor built around agents, often compared with Hermes.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
A diagramming tool to plan an agent before building it.
In your videos: Named in 1 of your videos
Watch it here:
Turns a prompt into a web app, with MCP connectors.
In your videos: Named in 3 of your videos
Watch it here:
Built-in skills that pass coding jobs to Claude Code or Codex, only when you ask.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Runs Claude Code tasks on Anthropic's machines, and pulls the session back when you want.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Keeps looping until a condition is met, and needs proof shown in the chat.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Small scripts that run on events in Claude Code to check work or add context.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 3 of your videos
Watch it here:
A built-in command that reviews your recent Claude Code sessions.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Add-ons that change how Claude Code looks and behaves.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Mods are temporary until you ask Claude to save one as a plugin
In your videos: Named in 1 of your videos
Watch it here:
Anthropic's official repository of Claude Code plugins.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 44.4k stars
Watch it here:
Changes how Claude Code shows its tool calls, with seven colour themes to pick from.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Runs several coding agent sessions side by side in the terminal.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Turns your code into a lasting knowledge graph so the agent uses fewer tokens exploring.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claims 99% (Sharbel) or 120x (Teku AI) fewer tokens; about 45k stars · Named in 4 of your videos
Popularity: 47.2k stars
Watch it here:
Meta's coding models, including small versions you can run locally.
In your videos: Named in 2 of your videos
Watch it here:
A settings panel and a manual fork that shares the cache.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Feeds the agent current documentation for code libraries.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 63.7k stars
Watch it here:
Saves space in Claude Code's working memory so long sessions last longer.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Nate Herk says skip it: usage-limit bug
In your videos: Named in 1 of your videos
Watch it here:
A very large open model made for coding.
About 18 skills that send tasks to coding agents you have installed, like Claude Code, Codex and Cursor.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A file view with a side-by-side site preview where you click parts to leave comments.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An autonomous coding agent, listed inside Kylon.
In your videos: Named in 1 of your videos
Watch it here:
Official Expo skill that helps Claude build mobile apps.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Adds a file panel on the right that shows which files Claude touched and which are saved.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Anthropic skill that avoids generic fonts and forces a clear design direction.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
A terminal program that Alex Finn uses.
In your videos: Named in 1 of your videos
Watch it here:
AI coding helper, listed among coding tools.
In your videos: Named in 1 of your videos
Watch it here:
Plays your Git history back as an animated replay.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Separate git worktrees let several agents work on one repo at the same time.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Guidelines from JetBrains that make Claude Code or Codex write up-to-date Go code.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Google's kit for building agents. It supports open models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Google's game builder.
In your videos: Named in 1 of your videos
Watch it here:
Builds a regex pattern for you from example strings.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Garry Tan's chain of role skills for planning, reviewing and shipping software.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: AI Edge cites 133k stars · Named in 3 of your videos
Popularity: 137.3k stars
Watch it here:
Coding agent repo that reviews its own past sessions. It may be the same Hindsight.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A framework for building agents and apps on language models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Open framework for building agents as connected graphs of steps.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A framework for building apps and agents that use your data with language models.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
An app builder for coding by vibes.
In your videos: Named in 2 of your videos
Watch it here:
A library of app screens and flows to copy design ideas from. See the Mobbin MCP.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Python kit for building teams of agents, shown working with Ollama models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A copy of Claude Code that can run other models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Presenters say its legality is unsettled
In your videos: Named in 1 of your videos
Watch it here:
Free, open-source terminal coding agent that works with local and hosted models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claimed 220k stars; already among Chris's built-in Hermes skills · Named in 6 of your videos
Popularity: 221.5k stars
Watch it here:
and 2 more videos
Blocks edits until you approve a plan.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Simon Scrapes says the Claude Code team is considering removing it
In your videos: Named in 1 of your videos
Watch it here:
A Microsoft tool that opens a real browser to test your app. Lighter than the MCP version.
Suits: Cyborg, Forge. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches; Forge could also use it.
In your videos: Named in 1 of your videos
Popularity: 98.3k stars
Watch it here:
Manages the add-on packages a Python project depends on.
In your videos: Named in 1 of your videos
Watch it here:
Applies about 20 design principles to web pages.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Gives you ready-made project templates on a Mac so you do not start from nothing.
In your videos: Named in 1 of your videos
Watch it here:
Runs attack and repeat tests on your prompts to catch weak spots and regressions.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Popularity: 24.2k stars
Watch it here:
Manages which Python versions are installed on your machine.
In your videos: Named in 1 of your videos
Watch it here:
Coding models you can run on your own machine.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Gives Claude the look and style of known apps, so what it builds follows that design.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Save a session under a name and bring it back after clearing.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Sample add-on with a replay button that steps through each edit Claude just made.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
An agent studies an app and rebuilds a feature or the whole app.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claimed 240k to 300k stars; named in seven videos · Named in 7 of your videos
Watch it here:
and 3 more videos
Undo recent turns without losing the prompt cache.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
VS Code add-on that works as an AI helper and can use MCP tools.
In your videos: Named in 1 of your videos
Watch it here:
A Claude Code add-on with a progress bar and a live panel of helper agents, models and cost.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
200 free animated website backgrounds that agents can use.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Local memory for coding agents, based on evidence from your project.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Database, login and backend memory for apps your agent builds. Has an official skill.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Nate Herk says set it read-only
In your videos: Named in 3 of your videos
Popularity: 107.9k stars
Watch it here:
A set of 20+ skills for planning, building, testing, reviewing and fixing code; also has a Hermes plugin.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Nate Herk still recommends it; Teku AI cites about 287k stars · Named in 5 of your videos
Popularity: 303.9k stars
Watch it here:
and 1 more videos
Gives agents current Swift and SwiftUI documentation for building Apple apps.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Breaks a product requirements document into tasks, using 36 MCP tools.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Turns a prompt into STL files for 3D printing.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Nuno Tavares says skip it: blocked on Windows 11
In your videos: Named in 2 of your videos
Watch it here:
A game engine that an agent can drive or import generated worlds into.
In your videos: Named in 2 of your videos
Watch it here:
A fast tool for starting, adding to and running Python projects.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 90.5k stars
Watch it here:
A terminal built for coding with AI help.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Builds calculators and dashboards from a plain description.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
An AI code editor that can use MCP tools.
In your videos: Named in 1 of your videos
Watch it here:
Manages Git worktrees, which suits running several Claude Code agents at the same time.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Reads, edits and converts YAML, JSON, XML, CSV and TOML files.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 15.9k stars
Watch it here:
The main program that runs your always-on agents. It has memory, skills, scheduled jobs and over 20 chat channels.
Suits: Alfred, Brainiac, Cerebra. It is the runtime all seven of Chris's agents already use.
In your videos: Alex Finn calls it the greatest AI tool ever made · Named in 56 of your videos
Popularity: 253.6k stars
Watch it here:
and 52 more videos
A chat app where your Hermes agents live, with one channel per project.
Suits: Alfred. Chris's agents already talk in Discord channels.
In your videos: Named in 12 of your videos
Watch it here:
and 8 more videos
A web page to see and manage your Hermes sessions, models, jobs, skills, keys, channels and logs.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
One strong model plans and hands work to cheap models, each given only the tools it needs.
Suits: Forge. Forge sets models; the doc notes Chris's agents all run Haiku, so this is the part not yet used.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
A health-check command that reports whether your Hermes install is working.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Agents save long reports in a documents area, not chat, which saves tokens.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
279 ready-made role-based agents across company divisions.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: AI Edge cites 152k stars · Named in 1 of your videos
Watch it here:
A live building picture showing which agents are working or idle.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Compare an agent's output with a quality bar or SOP kept in skills.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Dan Martell's steps for a first agent: Aim, Give identity, Equip, Narrow, Trust, with staged trust.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Paid dashboard that watches Claude, Codex, Hermes and Antigravity agents, with steer and stop buttons.
Watch out: Sold inside a paid community
In your videos: Named in 3 of your videos
Watch it here:
A desktop tab that collects every link, image and file you sent the agent.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Runs a second task in the background at the same time.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Optional hand-gesture 3D screen for your files, part of the Jarvis project.
In your videos: Named in 1 of your videos
Watch it here:
Block's team workspace that feels like Slack, where agents are members. You can host it yourself.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claimed 35,000 stars · Named in 3 of your videos
Watch it here:
Always-on OpenAI agent in the ChatGPT desktop app that can watch over Codex threads.
In your videos: Alex Finn calls it incredible · Named in 5 of your videos
Watch it here:
and 1 more videos
ChatGPT's agent mode with a computer, files and browser. It runs scheduled cloud workflows.
In your videos: Alex Finn says it has the best computer use · Named in 3 of your videos
Watch it here:
Beta: one main conversation that starts cloud threads.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A no-code agent inside Claude Desktop with plugins, connectors, scheduled tasks and computer control.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Tina Huang calls it the safer agent she lets read her email · Named in 6 of your videos
Watch it here:
and 2 more videos
An always-on Anthropic agent on Opus 5.5 with its own cloud computer, channels and to-do list.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Waitlist at the time
In your videos: Named in 3 of your videos
Watch it here:
Checks Hermes for clutter in its files, memory and tools, and sets items aside instead of deleting.
Suits: Forge, Sentinel. Forge runs agent ops and Sentinel reviews skills, so both benefit from a bloat audit that archives instead of deleting.
In your videos: Featured as the headline skill; the playlist-1 doc made it a top pick · Named in 1 of your videos
Watch it here:
A custom Claude Code command that connects to your server to edit and manage a Hermes agent.
Suits: Forge. Forge could manage the Hermes agents from Claude Code the same way.
In your videos: Named in 1 of your videos
Watch it here:
Runs dozens of agents at once from one script. Set the size small in the settings.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Lets an agent operate apps on your Mac.
In your videos: Named in 1 of your videos
Watch it here:
xAI's agent app. Each bot gets its own cloud computer, routines and reports.
In your videos: Named in 10 of your videos
Watch it here:
and 6 more videos
A shared chat for several bots, with a turn limit before you must step in.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Writes a short handoff note of a session so another agent can carry on.
Suits: Alfred, Forge. Alfred and Forge both hand work between agents and surfaces.
In your videos: Named in 3 of your videos
Watch it here:
Runs several specialist bots, each with its own personality, skills, memory and model, that can mention each other.
Suits: Forge. Forge runs the agent team; bot mode is how the seven agents could share rooms.
In your videos: Sharbel calls it a cheat code; Wanderloots suggests keeping 3 to 5 bots warm and a 10 to 30 minute idle timeout · Named in 3 of your videos
Watch it here:
Opens the Hermes settings to edit, for example to limit admin commands to one person.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 1 of your videos
Watch it here:
A desktop app for using Hermes: chats, models, profiles, scheduled jobs, skills and pinned sessions.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 9 of your videos
Watch it here:
and 5 more videos
Restarts the gateway so your config and channel changes take effect.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Ways to connect the Hermes app to an agent: local, Hermes Cloud, remote gateway or SSH.
Suits: Forge. Forge runs the Coolify-hosted agents the app would connect to.
In your videos: Teku AI uses the remote gateway to reach a VPS · Named in 1 of your videos
Watch it here:
Keeps the agent working until your stated goal is done, with a limit on turns.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Jack Roberts says a goal has a 20-turn budget by default
In your videos: Sharbel says to have Hermes write the goal prompt first (meta-prompting) · Named in 5 of your videos
Watch it here:
and 1 more videos
A task board where work moves from new to done and gets shared out between helper agents.
Suits: Forge, Brainiac. Forge runs builds and ops; Brainiac could own filing tasks on the board.
In your videos: Named in 10 of your videos
Watch it here:
and 6 more videos
Jack Roberts' set-up: a team of personas over Hermes, each with its own job, prompt and model.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Separate Hermes setups, each with its own keys, memory, scheduled jobs, skills and personality.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Alex Finn recommends at least two profiles so one can fix the other · Named in 6 of your videos
Watch it here:
and 2 more videos
Turns tools such as browser automation or computer use on or off for each platform.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Sharbel's mission-control dashboard for agents, hosted on Vercel and gated by Google sign-in.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
A workspace where a team of agents works inside Slack. Sign in with a ChatGPT plan.
In your videos: Named in 3 of your videos
Watch it here:
Open-source Claude Code assistant with voice, a visualizer, a persona and memory kept in an Obsidian vault.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Slack-style workspace where people and Claude Code or Codex agents share channels and automations.
In your videos: Matt Wolfe found it impressive; Nate Herk notes it was sponsored · Named in 2 of your videos
Watch it here:
Karpathy's project where models with different personas debate to improve a decision.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 25.6k stars
Watch it here:
A web agent that does tasks on its own.
In your videos: Named in 1 of your videos
Watch it here:
A custom dashboard your agent builds to show tasks, content, memory, docs and agent status.
Suits: Cyborg. Cyborg runs the LAN dashboard Chris is planning.
In your videos: Named in 4 of your videos
Watch it here:
A dashboard for several Hermes agents showing tasks, health checks and live costs.
Suits: Cyborg, Forge. Cyborg runs the dashboard side and Forge runs fleet ops; matches Chris's planned agent dashboard.
In your videos: Sharbel presents it as his own tool for running a Hermes fleet · Named in 2 of your videos
Watch it here:
Several named agents with set jobs, each on its own channel and model.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 4 of your videos
Watch it here:
A task board and runtimes for agents. Setup uses API keys and hooks.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A local agent framework. The catalogue only names it.
Starts a fresh session when the chat gets too full, which keeps costs down.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Greg Isenberg's guest says one session per topic keeps context slim · Named in 3 of your videos
Watch it here:
Splits a task between specialist agents, runs them side by side and checks the result.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Popularity: 40.2k stars
Watch it here:
Local agent platform you install with one command. It includes a security audit command.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Alex Finn suggests it as a backup agent; Greg Isenberg's guest switched away from it · Named in 18 of your videos
Popularity: 393.0k stars
Watch it here:
and 14 more videos
Asks Hermes during install to bring in an existing OpenClaw setup.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
Lets you run several coding agents in one window.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Runs Hermes, Claude Code, Codex and Pi agents like a company, with an org chart, tasks and secrets.
Suits: Forge, Cyborg. Forge runs agent ops and Cyborg dashboards; Paperclip can hire Hermes agents into one board.
In your videos: NetworkChuck says try it right now; claimed 100k+ stars · Named in 5 of your videos
Popularity: 105.5k stars
Watch it here:
and 1 more videos
Perplexity's agent, which now runs fully on a Mac.
Watch out: Better Creating calls it expensive; about 36 GB RAM suggested
In your videos: Named in 2 of your videos
Watch it here:
Unreleased local app with workspaces, sessions, bots, automations and dashboards over Claude Code or Codex.
In your videos: Named in 1 of your videos
Watch it here:
A very small agent harness that works well with local models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Popularity: 117.8k stars
Watch it here:
Lines up a prompt for the agent's next turn.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Mac app that puts several agents in one conversation with no API key.
In your videos: Named in 1 of your videos
Watch it here:
Clears the session or summarises it so the context gets smaller.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Sets your home channel for scheduled results and checks whether you are the admin.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
Builds an agent without code from your PDFs and pages, and can be put on a website.
In your videos: Named in 1 of your videos
Watch it here:
Team chat that agents can use, and a source for capturing notes.
In your videos: Named in 8 of your videos
Watch it here:
and 4 more videos
Opens sessions side by side, with one box able to drive many.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Lets you nudge the agent while it works without stopping it.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Stops the current task or ends background processes.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Small helper agents that the main agent starts for narrow jobs, then checks their work.
Suits: Forge, Cerebra. Forge and Cerebra can split work into parallel helpers.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
A desktop view to watch, stop or steer each running helper agent and change its model.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
Breaks a big goal into 4 to 10 actions shown on a dashboard.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A control screen and phone app for Claude Code, Codex and OpenCode.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Nimrod uses it to run 100+ projects · Named in 2 of your videos
Watch it here:
The most recommended phone chat for Hermes, with BotFather bots, QR setup and group topics as rooms.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Alex Finn recommends it for deep mobile work · Named in 24 of your videos
Watch it here:
and 20 more videos
A messaging channel for Hermes and OpenClaw.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
A file that tells each agent who it is and how to behave.
Suits: Forge. Forge maintains the agents' own files.
In your videos: Jack Roberts calls it a character bible and gives a template · Named in 10 of your videos
Watch it here:
and 6 more videos
A file of instructions Claude reads. Keep it a short map to other files, and check it now and then.
Suits: Forge, Brainiac. Forge builds and runs agent operations, where this belongs; Brainiac could also use it.
Watch out: Mason Fitzgerald calls it a soft instruction that is not enforced
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
One shared instruction file for agents. Claude Code now reads it, and it can sit in any folder.
Suits: Forge, Brainiac. Forge builds and runs agent operations, where this belongs; Brainiac could also use it.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Your agent saves repeated jobs, checklists or SOPs as its own reusable skills.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
A team folder of markdown skill files that every agent reads.
Suits: Forge, Brainiac. Forge builds and runs agent operations, where this belongs; Brainiac could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Try a job with no skill first, and write down only what the model gets wrong.
Suits: Forge, Sentinel. Forge and Sentinel keep skills small, matching the 15 KB guideline from video 6.
In your videos: Named in 1 of your videos
Watch it here:
A site that explains what agent skills are.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A curated list of Claude skills.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Popularity: 15.5k stars
Watch it here:
Record yourself doing a job, then turn it into an agent playbook.
In your videos: Named in 1 of your videos
Watch it here:
A marketplace of skills, agents, MCPs and plugins for Claude Code.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Popularity: 30.8k stars
Watch it here:
Helps Claude write documents together with you.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
A big bundle of 292 skills that guide Claude through plan, test, build, review and check steps.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Turns a PDF, link or YouTube video into a new saved skill for your agent.
Suits: Brainiac, Cerebra. Brainiac and Cerebra could turn sources into reusable skills.
In your videos: Named in 2 of your videos
Watch it here:
A screen to browse official and community Hermes plugins with badges and reviews.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 3 of your videos
Watch it here:
Rewrites text to remove the usual signs that an AI wrote it.
Suits: Alfred, Mantis. Alfred and Mantis write text for Chris; it is already built into Hermes.
Watch out: Teku AI says it is token heavy; Sharbel notes the skill file is 29.6 KB
In your videos: Playlist-1 doc notes it is already built into Hermes · Named in 4 of your videos
Popularity: 60.1k stars
Watch it here:
Gives Claude 24 design commands, like critique and delight, to improve how web pages look.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
A prompt structure: Mission, Ask, Parameters, Shape.
In your videos: Named in 1 of your videos
Watch it here:
A pack of 15 to 27 coding skills, including Grill Me, which asks questions before acting.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 5 of your videos
Popularity: 302.9k stars
Watch it here:
and 1 more videos
Ask a model to write the agent's prompt or goal before you run it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Skill pack from Obsidian's CEO that teaches agents the Obsidian command line, markdown, Bases, Canvas and Defuddle.
Suits: Brainiac. Brainiac writes vault notes and needs correct wikilinks and front matter.
In your videos: Named in 2 of your videos
Popularity: 50.0k stars
Watch it here:
Anthropic's reference examples of skills.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Write one skill per project and keep adding context to it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Switches the persona for one session without changing the SOUL.md file.
Suits: Mantis, Alfred. Mantis does image prompts and creative work, which this produces; Alfred could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Keeps plan, findings and progress files so the agent does not lose its plan.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 27.5k stars
Watch it here:
A Claude desktop feature that watches you do a job and writes a small skill from it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Saves lessons from a bot chat into its skills or memory.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Turns your corrections into a saved rule in CLAUDE.md or a skill.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Redesign skill that reads your files to learn how your design is put together.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Dump your context, then let the agent interview you or write the best prompt itself.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 3 of your videos
Watch it here:
One big skill runs smaller skills in order, such as microsite, then copy, then QA.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
After each session, cleans up and improves the agent's skill collection.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An Anthropic helper skill that builds and improves new skills.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Writes new skills and checks whether an existing skill should be extended instead.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Catalogues of shared agent skills you can browse.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
Watch out: Tina Huang avoids ClawHub skills unless from trusted authors
In your videos: Named in 2 of your videos
Watch it here:
Runs a job with and without a skill to check whether the skill really helps.
Suits: Sentinel, Forge. Sentinel and Forge could prove a skill earns its place before it ships.
In your videos: Named in 1 of your videos
Watch it here:
Hermes catalogue of community skills you can install.
Suits: Sentinel. Sentinel should vet anything installed from the hub.
Watch out: Metics Media says to review a skill before use
In your videos: Named in 2 of your videos
Watch it here:
Website design skill you install from GitHub into Codex or Claude.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
A skill that finds the best skill for your task, listed on skills.sh.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Popularity: 36.3k stars
Watch it here:
A doc built by interview about how you think, plus a voice guide per medium with banned words.
Suits: Alfred, Mantis. Alfred, the general assistant, could use it in everyday chats; Mantis could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Anthropic add-on where a second agent points out things you may have missed.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
Watch out: Uses extra tokens
In your videos: Named in 1 of your videos
Watch it here:
Note app using plain markdown files. Your agents use it as their memory, wiki and workspace.
Suits: Brainiac, Cyborg, Cerebra. Brainiac files the vault, Cyborg publishes it and Cerebra indexes it.
Watch out: Mason Fitzgerald argues file search over Obsidian is slow and costly versus RAG
In your videos: Named in 20 of your videos
Watch it here:
and 16 more videos
Two notes files Hermes reads at the start of every session so it remembers you.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Metics Media had the agent build a NOTES_INDEX.md as well · Named in 9 of your videos
Watch it here:
and 5 more videos
A searchable wiki of decisions, logs and research that your agent keeps up to date so it stays consistent.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 4 of your videos
Watch it here:
Each agent writes down what it did, or adds it to today's daily note, before replying.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 3 of your videos
Watch it here:
Ask for a handoff file at around 200 to 300k tokens, then restart in a fresh chat.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Saves web pages and YouTube transcripts into your vault's raw folder.
Suits: Cerebra, Brainiac. Cerebra does web and source-graded research and vault indexing, which this feeds; Brainiac could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Export what your AI system knows about you as markdown, to set up a new agent.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
A folder of business markdown docs that SOUL.md tells the agent to read.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Tell the agent to write everything it does into daily logs.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Local workspace where you upload documents and run agent workflows.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Dan Martell's second-brain product.
In your videos: Named in 1 of your videos
Watch it here:
Agent memory kept as plain markdown files with wiki links. It replaces Notion.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
Watch out: Hosted plan costs $15 a month
In your videos: Named in 1 of your videos
Popularity: 3.6k stars
Watch it here:
Folders of markdown notes with README files that agents search, read and write back to.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A docs-like space in ChatGPT where Dots logs milestones and bugs.
Keeps a lasting memory for Claude Code so it remembers things between sessions.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
Watch out: Nate Herk says skip it: it sends data by default
In your videos: Named in 1 of your videos
Watch it here:
Community Obsidian plugin for chatting with AI about your vault.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Agent memory system and code analyzer that works with Graphify.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Scheduled scripts write numbers into daily-note headers so your notes work like a table you can query.
Suits: Brainiac, Cyborg. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cyborg could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Each workflow folder holds lessons learned, which the agent reads before every run.
Suits: Brainiac, Forge. Brainiac files and maintains the vault and runs cron jobs, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Watch it here:
A tool that builds a knowledge graph. Named as Cortex Suite's partner.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Reviews recent activity across your agent tools and gathers the lessons learned.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Adds one outside memory service, such as Honcho, Mem0 or Supermemory, to Hermes.
Suits: Brainiac, Cerebra. Brainiac owns memory and Cerebra indexes the vault.
Watch out: Sharbel says only one external provider can be active at a time
In your videos: Named in 2 of your videos
Watch it here:
Long-term agent memory that can retain, recall and reflect. A Hermes memory option that runs in Docker.
Suits: Brainiac, Cerebra. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cerebra could also use it.
Watch out: Playlist-1 doc notes it overlaps Chris's Qdrant plan
In your videos: Named in 4 of your videos
Popularity: 44.6k stars
Watch it here:
Memory service shared across agents, with dreaming. Check it with the command hermes honcho status.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 4 of your videos
Popularity: 7.8k stars
Watch it here:
The same repo on every machine lists your projects, so threads can start remotely.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 1 of your videos
Watch it here:
A rule that agents may not edit the human side of your vault. The machine side is theirs.
Suits: Brainiac, Sentinel. Brainiac and Sentinel already guard off-limits folders like the archive.
In your videos: Named in 1 of your videos
Watch it here:
Agent framework with memory built in.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Hermes skill where a librarian bot files reports into your Obsidian wiki and asks you to approve reviews.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 2 of your videos
Watch it here:
Shared memory across Hermes, Claude Code and Codex through MCP. It has a free tier.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 2 of your videos
Popularity: 67.3k stars
Watch it here:
File graph like Obsidian combined with RAG. The name may be garbled.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A workspace for SOPs and task pipelines, also used as an alternative home for agent memory.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 10 of your videos
Watch it here:
and 6 more videos
Notion's built-in AI, described as a locked-in black box compared with building your own brain.
In your videos: Named in 1 of your videos
Watch it here:
Database-style views over your vault notes.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A visual board in Obsidian for charts, boards and views.
Suits: Brainiac, Mantis. Brainiac files and maintains the vault and runs cron jobs, which this supports; Mantis could also use it.
In your videos: Named in 2 of your videos
Watch it here:
An Obsidian setting that lets an agent work on your vault directly.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Custom Obsidian plugin with focus, metrics, a heat map and a quick-capture inbox.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
Watch out: Sold through his community
In your videos: Named in 1 of your videos
Watch it here:
A built-in skill that saves and reads notes in your Obsidian vault as extra memory.
Suits: Brainiac. Brainiac files notes into the vault.
In your videos: Named in 1 of your videos
Watch it here:
Paid, encrypted sync so several computers and agents share one vault.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A beta switch that moves strong memories into memory.md.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Notes app built on Emacs Org files, with a web version you can host yourself.
In your videos: Named in 1 of your videos
Watch it here:
Stores your documents, adds tags by rules, and lets you search them.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A vector database included in the n8n self-hosted AI starter kit, for searching by meaning.
Suits: Cerebra. Cerebra's vault semantic index is planned on Qdrant.
In your videos: Named in 1 of your videos
Popularity: 34.9k stars
Watch it here:
A vector database plus a hook that adds the 8 closest notes before Claude sees your prompt.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: Package sits behind a paid community; needs Claude Code, not Cowork or web
In your videos: Claims RAG is 5.2x faster and 3.5x cheaper than file search · Named in 1 of your videos
Watch it here:
Searches your old past conversations with the agent so you can find what was said.
Suits: Alfred, Brainiac. Alfred and Brainiac can recall earlier chats on demand.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
A short CLAUDE.md map plus index files, built from an LLM interview, so agents find context fast.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Pav Rusovs measured 30s vs 62s and 17k vs 28k tokens with the map · Named in 1 of your videos
Watch it here:
Community Obsidian plugin that links AI to your vault.
Suits: Brainiac, Cerebra. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cerebra could also use it.
In your videos: Named in 1 of your videos
Popularity: 5.5k stars
Watch it here:
Fast cloud memory that several agents can share.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 2 of your videos
Watch it here:
OpenAI model that turns notes into searchable numbers. Suggested for a vault search, about $5 credit.
Suits: Cerebra. Cerebra's semantic vault index needs an embedding model.
In your videos: Named in 1 of your videos
Watch it here:
Keep decisions and working docs, then turn them into lessons in the brain.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Pav Rusovs' skills: nightly map rebuild, broken-link check, Mac cleanup and a media skill.
Suits: Brainiac. Brainiac maintains the vault and could lint broken links nightly.
In your videos: Named in 1 of your videos
Watch it here:
Starter kit with a wiki_tool.py script, ingest, query, maintain and lint skills, and note templates.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
Watch out: Starter kit is for members only
In your videos: Named in 1 of your videos
Watch it here:
Instructions tell agents to read the wiki first and drop notes into an inbox that a script files.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Memory layer that tracks how facts change over time.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A notebook that answers from your own sources, and makes audio, video, slides, mind maps and reports.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Lets agents read sites like X, Reddit, YouTube and GitHub without paying for API fees.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Teku AI cites about 81k stars and says it gets past 403 bot blocks · Named in 4 of your videos
Popularity: 101.3k stars
Watch it here:
Strips a web page down to clean readable text, which saves on tokens.
Suits: Cerebra, Brainiac. Cerebra fetches pages for research and Brainiac clips into the vault; both save tokens.
In your videos: Named in 4 of your videos
Popularity: 9.4k stars
Watch it here:
Thousands of ready-made scrapers that agents can use.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 13.2k stars
Watch it here:
Karpathy's loop that runs experiments overnight while you sleep.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A bundled skill that gets a video transcript for summaries or scheduled quizzes.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
An open-source web scraper, the second choice after the main one.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A scheduled job scans Reddit and X for business ideas and builds a dashboard.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Multi-agent research inside Claude Code that cross-checks its findings.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A multi-step research process that uses several models at once.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
Company-wide search plus an agent layer.
In your videos: Named in 1 of your videos
Watch it here:
A browser console showing live plane, ship and satellite data.
In your videos: Named in 1 of your videos
Watch it here:
An agent that researches a topic on its own and writes a report.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 29.5k stars
Watch it here:
Adds footnote-style sources to research results so you can see where facts came from.
Suits: Cerebra. Cerebra's source-graded research needs citations.
In your videos: Named in 1 of your videos
Watch it here:
Researches a topic from the last 30 days across Reddit, X, YouTube, GitHub and the web.
Suits: Cerebra. Cerebra does web research and this gives her fresh multi-platform sentiment.
In your videos: AI Edge notes 140 contributors; Futurepedia lists it first · Named in 3 of your videos
Popularity: 65.2k stars
Watch it here:
Connects Hermes to NotebookLM for research and a podcast-style daily brief.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
An open-source skill from Sharbel for researching YouTube content, loaded automatically.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
OpenAI reasoning models used to tidy up research questions before searching.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
An AI tool for looking up news and researching topics.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A model that searches the web as it answers. Used by agents for web research.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 3 of your videos
Watch it here:
Web research tool you add to n8n agents.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
Chat with your own PDFs on your computer, at localhost:8001.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
More than 160 skills for science research tasks.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: Video 17 reports one science skill pushes a paid product through its skill file
In your videos: Named in 2 of your videos
Watch it here:
API that scrapes social media and YouTube.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Python scraper with a stealth fetcher. Offered as a free Firecrawl replacement and optional Hermes skill.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: Nuno Tavares says skip: it gets around blockers
In your videos: Claimed 87,000 stars · Named in 2 of your videos
Popularity: 87.9k stars
Watch it here:
Google's model for forecasting numbers over time.
In your videos: Named in 1 of your videos
Watch it here:
Replaces the default YouTube skill with transcripts, channels and search, with no API key.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Turns YouTube videos into markdown notes using the Karpathy wiki method.
Suits: Cerebra, Brainiac. Cerebra does web and source-graded research and vault indexing, which this feeds; Brainiac could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Runs a task on a timer by itself and sends the result to a chat channel.
Suits: Brainiac. Brainiac runs Chris's cron work.
In your videos: Alex Finn says to prune and pause old cron jobs to improve speed and cost · Named in 13 of your videos
Watch it here:
and 9 more videos
Self-hosted workflow automation for no-code agents, with Evaluations nodes and a starter kit.
Suits: Brainiac, Cerebra. Chris runs n8n for the nightly note and research workflows.
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
A daily scheduled job that asks your top priority, briefs you and starts work.
Suits: Alfred, Brainiac. Alfred, the general assistant, could use it in everyday chats; Brainiac could also use it.
In your videos: Named in 4 of your videos
Watch it here:
The agent writes email replies in your voice but never sends them on its own.
Suits: Alfred, Sentinel. Alfred, the general assistant, could use it in everyday chats; Sentinel could also use it.
In your videos: Named in 1 of your videos
Watch it here:
A database for research sections and citations inside n8n.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
An hourly routine gathers Slack, meetings and email into an inbox, then a librarian step files it.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Cloud jobs that start on a schedule, an API link with a token, or a GitHub event.
Suits: Brainiac, Cyborg. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cyborg could also use it.
In your videos: Greg Isenberg's guest calls it Cowork scheduled tasks on steroids · Named in 2 of your videos
Watch it here:
Lets scheduled jobs remember earlier runs so they do not repeat work.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A scheduled-job option that runs only a script, saving tokens by skipping the AI.
Suits: Brainiac. Brainiac's scheduled vault jobs could run as plain scripts.
In your videos: Named in 1 of your videos
Watch it here:
Double-click commands that draft with a local model into a review folder before anything is applied.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Visual builder where you draw an agent pipeline instead of writing code.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
After each call, Hermes pulls the transcript and turns action items into tasks.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Nodes that score your agent's answers against test cases kept in Google Sheets.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A bundle of n8n with Ollama, Qdrant and Postgres, ready to self-host.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 15.2k stars
Watch it here:
Lets a connected app trigger the agent through MCP events.
In your videos: Named in 1 of your videos
Watch it here:
A YAML list of routines checked every 5 minutes with a run cap, plus a read-only dashboard.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
An agent watches sign-ups and pages through webhooks and controls payments.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Do actions across Zapier's 9,000 apps by chatting, with one central log.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 4 of your videos
Watch it here:
Reusable scripts with saved sign-ins. Cheaper than MCP for repeated tasks.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Corbin claims about 210x cost cut versus MCP in a separate video · Named in 3 of your videos
Watch it here:
Turns websites into clean markdown. It can scrape, crawl and get past bot detection.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: The Next New Thing offers Scrapling as a free replacement
In your videos: Named in 4 of your videos
Popularity: 194.9k stars
Watch it here:
Android Debug Bridge, used to connect an Android phone to Hermes.
In your videos: Named in 1 of your videos
Watch it here:
Hosted browser plugin so Hermes can automate the web.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Connects an agent to your real Chrome browser and fixes itself when steps fail.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 18.3k stars
Watch it here:
Browser automation plugin for Hermes.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 119.1k stars
Watch it here:
Plugins linking ChatGPT or Codex to Gmail, Notion, Slack, Drive, Calendar, Supabase and Netlify.
In your videos: Named in 2 of your videos
Watch it here:
Chrome extension that lets ChatGPT or Codex control your browser and computer.
In your videos: Named in 3 of your videos
Watch it here:
Builds a command-line tool for any desktop app so agents can drive it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Lets Codex read the iMessage database stored on your Mac.
In your videos: Named in 1 of your videos
Watch it here:
Connects agents to over 1,000 apps like Gmail, Slack and Notion through sign-in, and has a command line.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 5 of your videos
Popularity: 30.7k stars
Watch it here:
and 1 more videos
Python kit for building your own MCP servers and clients.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 24.3k stars
Watch it here:
Lets an agent use Drive, Gmail, Calendar, Docs, Sheets and Slides, with over 40 skills.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Popularity: 30.4k stars
Watch it here:
Switch-on connections inside Hermes to tools like Canva, Notion, GitHub and NotebookLM.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Tests MCP servers. Run it with npx @modelcontextprotocol/inspector.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 10.7k stars
Watch it here:
A list with switches to turn off MCP servers you do not use.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Makes agent replies appear as visual screens like dashboards and forms.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Claimed 13,000 stars · Named in 1 of your videos
Watch it here:
An open chat interface for AI models that plugs into n8n and MCP connectors.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A community plugin that lets Hermes run approved scripts on a real Android or mirrored iPhone.
Suits: Sentinel. Sentinel would need to lock it down before any use.
Watch out: On by default in messaging apps, so others messaging the bot could control the phone; needs Hermes 0.21.6 or newer
In your videos: Named in 1 of your videos
Watch it here:
Free iMessage link for agents, with no Mac mini needed.
In your videos: Named in 1 of your videos
Watch it here:
Microsoft's official tool for controlling a browser using page accessibility snapshots.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Popularity: 38.0k stars
Watch it here:
Adds a real web browser beside the chat, including GitHub pull requests.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
API for LinkedIn and Instagram DMs, WhatsApp, Messenger, Telegram and email, about $5 per account.
In your videos: Named in 1 of your videos
Watch it here:
Send voice notes from your phone that an agent on a home machine turns into text and acts on.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Standalone MCP server that works without Zapier, using your own tokens.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A private, encrypted network so your agents can reach and log into all your devices.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 8 of your videos
Popularity: 36.3k stars
Watch it here:
and 4 more videos
Saves versions of your agent or vault, with .gitignore, and backs it up daily to GitHub.
Suits: Cyborg. Cyborg already publishes the vault to GitHub.
In your videos: Named in 3 of your videos
Watch it here:
A scheduled job that checks your agent's security every hour or twice a day.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 3 of your videos
Watch it here:
Makes the agent ask you first before it spends money, sends things or deletes anything.
Suits: Sentinel, Forge. Sentinel enforces safety and Forge runs ops; gates match Chris's no-surprise rules.
In your videos: Named in 2 of your videos
Watch it here:
Checks a new skill for hidden bad instructions before you install it.
Suits: Sentinel. Sentinel already security-vets skills; Bouncer automates that first check.
In your videos: Named in 1 of your videos
Watch it here:
Never paste keys into chat. Set them as environment variables in the terminal and give least access.
Suits: Sentinel. Sentinel enforces Chris's no-secrets rule.
In your videos: Named in 1 of your videos
Watch it here:
Open the Hermes dashboard only through a VPN or SSH tunnel, never openly on the internet.
Suits: Sentinel, Cyborg. Sentinel reviews and security-vets skills, which this supports; Cyborg could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Give agents read-only keys and database users that cannot see tables holding personal data.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Pauses risky deletes, lists the affected files and waits for you to press Proceed or Cancel.
Suits: Sentinel. Sentinel would like a hard stop on deletes; it fits Chris's archive-not-delete rule.
In your videos: Named in 1 of your videos
Watch it here:
Free monitoring for Linux, Docker, Proxmox and websites, with checks for agent health.
Suits: Cyborg, Sentinel. Cyborg runs the LAN telemetry dashboard and Sentinel watches health; fits Chris's planned agent dashboard.
In your videos: NetworkChuck uses it to monitor his Hermes agents (crons, gateway, model login) · Named in 1 of your videos
Watch it here:
Anthropic's multi-agent review that checks your code for security problems.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 2 of your videos
Watch it here:
Manages your domains and subdomains. DDNS updates them when your public IP changes.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Turns a coding agent into a security auditor.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Makes a local model or app reachable from the internet.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
A server control panel template available on Hostinger.
In your videos: Named in 1 of your videos
Watch it here:
Safe, sealed-off workspaces from Dagger where an agent can work without touching your real machine.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Popularity: 4.0k stars
Watch it here:
Checks whether a DNS change has spread across public resolvers.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Runs several apps and agents separately on one machine. Docker Compose handles whole stacks.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Runs agent code and browsing inside Docker so it stays walled off from the rest of your server.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 2 of your videos
Watch it here:
A self-hosted panel to deploy apps, databases and stacks, with GitHub auto-deploy and previews.
Suits: Forge. Forge runs deploys; Dokploy is an alternative to Chris's Coolify.
In your videos: Named in 2 of your videos
Watch it here:
One safe place to store your API tokens. It syncs to Vercel and has a skill.
Suits: Sentinel. Sentinel would keep secrets out of chats and Discord.
In your videos: Named in 1 of your videos
Watch it here:
A skill that hosts and deploys websites your agent has made.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
A platform for deploying apps.
In your videos: Named in 1 of your videos
Watch it here:
A low-cost server provider, about 5 to 6 dollars a month.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Hetzner's command line plus an API token, so Claude Code can create and manage servers.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A rented server with one-click Hermes, Dokploy and n8n templates and the Kodee assistant.
Suits: Forge. Forge runs hosting, but Chris already hosts on Coolify, so this is only a comparison.
Watch out: Lead Gen Jay calls Hostinger servers weak
In your videos: Named in 12 of your videos
Watch it here:
and 8 more videos
Gives you free SSL certificates for your sites.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A self-built MCP that shows agents only the API routes you allow, such as Redmine.
Suits: Sentinel, Forge. Sentinel would want least-privilege API access for every agent.
In your videos: Named in 1 of your videos
Watch it here:
A settings file that blocks sending, sharing and deleting on the server. Email stays draft only.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Security review skills, said to be by Google, that can run on a schedule.
Suits: Sentinel. Sentinel does security review; note the name clashes with Chris's Mantis agent.
In your videos: Named in 1 of your videos
Watch it here:
Hosts websites, and works as a connector to publish SEO content.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Gives self-hosted apps proper domain addresses and fixes CORS errors.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A Linux system set up declaratively, using Nix flakes.
In your videos: Named in 1 of your videos
Watch it here:
Works out how to build an app just by looking at its source.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Script and skill that limit which vaults an agent is allowed to touch.
Suits: Sentinel, Brainiac. Sentinel and Brainiac could enforce the archive-folder ban with an allow-list.
In your videos: Named in 1 of your videos
Watch it here:
A text screen in your terminal for viewing and managing your Docker containers.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 1.9k stars
Watch it here:
A dashboard for Docker containers that deploys stacks from compose files.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A database service for self-hosted apps and n8n.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
Software for running virtual machines, named as a home for Paperclip or Hermes.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A platform for deploying apps.
In your videos: Named in 1 of your videos
Watch it here:
A cache used in Buzz's self-hosted stack.
In your videos: Named in 1 of your videos
Watch it here:
Detects deepfakes and AI-made content.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
When a Telegram bot is open to everyone, allow admin commands for one Telegram ID only.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Rent powerful graphics cards by the hour.
In your videos: Named in 1 of your videos
Watch it here:
Scans agent skills for security problems before you install them.
Suits: Sentinel. Sentinel vets skills, so a scanner is a direct fit.
In your videos: Named in 1 of your videos
Watch it here:
Links a few devices together, as a replacement for Tailscale.
A Tailscale beta feature that shares a folder as a network drive.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Quickly sends files between your Tailscale devices.
In your videos: Named in 1 of your videos
Watch it here:
Shares a local port to your own Tailscale network with a web address.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
A front door that gives the Hermes dashboard HTTPS in Hostinger's Docker template.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Basic server lockdown: a sudo user, ed25519 SSH keys, and a firewall allowing only SSH, 80 and 443.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Self-hosted analytics for seeing who visits your website.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
Watch out: Default login admin/umami must be changed
In your videos: Named in 1 of your videos
Watch it here:
Free hosting for dashboards and sites, with a Doppler sync.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 4 of your videos
Watch it here:
Claude plugin used in a demo to publish a design to Vercel.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
One API key that reaches hundreds of models, with spending limits per key, rankings and free models.
Suits: Forge. Forge manages models and each agent's OpenRouter key and budget.
In your videos: Moe Lueker uses it as a cost router to the cheapest capable model · Named in 14 of your videos
Watch it here:
and 10 more videos
Anthropic's fast, cheap model. Good for small coding helper agents where cost matters.
Suits: Alfred, Brainiac. All seven agents run on Haiku today.
In your videos: Sharbel suggests Haiku sub-agents to protect the main cache · Named in 8 of your videos
Watch it here:
and 4 more videos
Send each question to the smallest model that can handle it, to keep costs down.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Shows where your Claude tokens go and flags MCP servers you never use.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Lets a cheap model ask a stronger one, like Opus, when it gets stuck.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A browser playground and leaderboard for comparing models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Runs 100 answer attempts in a bracket and keeps the best.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A usage number shown in the corner of the Claude Code screen.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Warns you about the cost before you send a message once the prompt cache has gone cold.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An ultra-short answer mode that claims about 65% fewer output tokens.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Nate Herk's test found only about 8.5% savings
In your videos: Teku AI cites about 100k stars · Named in 3 of your videos
Popularity: 114.0k stars
Watch it here:
Anthropic's strongest tier, named as an advisor model above Opus.
Watch out: Jason Lee hit usage limits with it
In your videos: Named in 12 of your videos
Watch it here:
and 8 more videos
Anthropic's top model line. Opus 5.5 is the one Anthropic suggests as your default.
Suits: Forge. Forge could use it as a strong orchestrator; the playlist-1 doc notes Chris's agents all run Haiku today.
Watch out: Alex Finn says it costs him about 40 USD a day; Moe Lueker says avoid it for routine tasks
In your videos: Named in 24 of your videos
Watch it here:
and 20 more videos
Anthropic's middle model, called the consensus best choice to power Hermes.
Suits: Forge, Cerebra. Forge builds and runs agent operations, where this belongs; Cerebra could also use it.
In your videos: Dan Martell: Sonnet to run agents, Opus to build them · Named in 13 of your videos
Watch it here:
and 9 more videos
Resets the Claude Code context for free.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Shows what is filling the context window.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Shows what the current session has cost.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An open model built to reason step by step.
Cheap open models that use tools well. Popular choice to power Hermes.
Suits: Alfred, Cerebra. Alfred, the general assistant, could use it in everyday chats; Cerebra could also use it.
In your videos: Metics Media picks V4 Flash as a cheap main model · Named in 16 of your videos
Watch it here:
and 12 more videos
Backup models and accounts so your agent keeps working when one provider fails.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A local helper that cuts token costs by sending simple calls to a cheaper route.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Weak until trained
In your videos: Named in 2 of your videos
Watch it here:
Chains about 635 free AI providers together. Comes as a desktop and Android app.
Watch out: Nuno Tavares says skip: the free models are weak
In your videos: Named in 2 of your videos
Watch it here:
Google's model family, used as cheap writers and agent drivers.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: Samin Yasar says skip it
In your videos: Tina Huang likes Gemini Flash for writing into the vault · Named in 6 of your videos
Watch it here:
and 2 more videos
Google's top model, priced at $2 in and $10 out per million tokens for a limited time.
In your videos: Named in 1 of your videos
Watch it here:
Small AI models that run on the device itself, not in the cloud.
In your videos: Named in 1 of your videos
Watch it here:
Google's open model for phones and local use, including uncensored versions.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
An open model from Z.AI. It is the cheapest option Alex Finn names.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 9 of your videos
Watch it here:
and 5 more videos
A phone app to download and run Gemma 4 locally.
Free notebooks with a T4 GPU, used to run or fine-tune models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A small OpenAI model used to test n8n agents.
In your videos: Named in 1 of your videos
Watch it here:
ChatGPT-plan models you can use in Hermes by logging in, with no API cost.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Alex Finn says 5.5 is the first usable cheap option via a ChatGPT subscription · Named in 11 of your videos
Watch it here:
and 7 more videos
OpenAI's top model, strong at using computers and coding. It powers ChatGPT Dots.
In your videos: Alex Finn says only Opus 5.5 is ahead of it · Named in 13 of your videos
Watch it here:
and 9 more videos
OpenAI's light, cheap model tier. Good value for everyday agent work.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 4 of your videos
Watch it here:
OpenAI's open-weight model. The 120B size needs large local hardware.
In your videos: Named in 2 of your videos
Watch it here:
OpenAI's top tier, used for heavy monthly reviews.
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
OpenAI's middle-tier model.
In your videos: Named in 2 of your videos
Watch it here:
xAI's model, linked by X login. It can read and search X.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 8 of your videos
Watch it here:
and 4 more videos
A hosted playground and fast API for running AI models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Nous Research's open model family for running locally.
In your videos: Named in 1 of your videos
Watch it here:
Terminal commands to set up your AI providers and switch the model in use.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
A hub for open models and Spaces demos you can try.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
Tencent's open model, free on some providers.
In your videos: Named in 3 of your videos
Watch it here:
A cheap, fast model for making decisions and scoring. The name may be wrong.
In your videos: Named in 3 of your videos
Watch it here:
Moonshot's open model, used as a cheap option to power Hermes.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 11 of your videos
Watch it here:
and 7 more videos
Meta's open models, including large and uncensored versions.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Runs local models directly and faster. Ollama is built on top of it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Checks your computer's hardware and suggests local AI models that will fit.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A desktop app to download local AI models and chat with them offline.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Lower GPU offload and context length if it fails to load a model
In your videos: Named in 3 of your videos
Watch it here:
Finds your computer hardware and picks and loads the best local model for it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Uses a cheap model for background jobs while the main model handles your chat.
Suits: Forge. Forge sets models; this saves OpenRouter credit for all seven agents.
In your videos: Named in 1 of your videos
Watch it here:
Meta's flagship models, Muse Spark (open) and Muse Wonder.
In your videos: Named in 3 of your videos
Watch it here:
Models that answer with probabilities, to help make decisions.
In your videos: Named in 1 of your videos
Watch it here:
An open-weight base model with 80B active parts. Not available yet.
In your videos: Named in 1 of your videos
Watch it here:
An open model named as Tina Huang's favourite for Hermes.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Lets you train a tiny language model from scratch in PyTorch.
In your videos: Named in 1 of your videos
Watch it here:
Mistral's models, for hosted use or small local use.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Sends one question to several models, then a lead model merges their answers.
Suits: Cerebra. Cerebra could cross-check research answers across models.
In your videos: Named in 2 of your videos
Watch it here:
Changes which AI model Hermes uses, right from the chat.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 4 of your videos
Watch it here:
Shows how much of your AI subscription usage is left.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
A small language model.
In your videos: Named in 1 of your videos
Watch it here:
Nvidia open models, free on OpenRouter for small side jobs your agents do.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 4 of your videos
Watch it here:
A Nous Research subscription, about 20 dollars a month, giving Hermes 200+ models and hosted Hermes.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 9 of your videos
Watch it here:
and 5 more videos
Manages local models and serves them at localhost:11434. Works with Hermes and Ollama Cloud.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 9 of your videos
Popularity: 184.7k stars
Watch it here:
and 5 more videos
An OpenAI endpoint that returns simple choices, probabilities or scores instead of text.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A list of other open and business models from Tina Huang's tier tour.
In your videos: Named in 1 of your videos
Watch it here:
Open 4B and 27B models that make decisions.
In your videos: Named in 1 of your videos
Watch it here:
Skill that cuts token use. One tester measured about 10 percent saved.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Changing model, effort, MCPs, plugins, or compacting resets the cache and costs more.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A local open model that powers Tina Huang's Hermes on a Mac Studio.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Alibaba's open models in many sizes, made for running local agents.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 11 of your videos
Watch it here:
and 7 more videos
Lets you pick how hard the model should think, next to the model picker.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
An open-weight model mentioned in a news roundup.
In your videos: Named in 1 of your videos
Watch it here:
Filters terminal output before it reaches the agent's working memory.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Nate Herk's test found no savings; AI Labs' RTK skill is paid
In your videos: Named in 2 of your videos
Popularity: 84.0k stars
Watch it here:
Sends the model only the skills it needs, cutting skill text from about 11k to 2.3k tokens.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Tiny story models small enough to run on an ESP32-S3 chip.
Managed services that run open AI models for you through an API.
In your videos: Named in 1 of your videos
Watch it here:
Have the agent measure and cut its own usage on each task.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A paste-in prompt that checks Claude Code context, deferral, memory files, cache and schedules.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A default mode that loads only a tool index, roughly 85% cheaper.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Turns off tools and built-in skills you do not use, to cut cost and wrong tool loops.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 2 of your videos
Watch it here:
A cheap open-weight model at $0.15 in and $0.50 out.
In your videos: Named in 1 of your videos
Watch it here:
Fine-tunes open models, either on your own computer or in Colab.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A model option you can use in Hermes.
In your videos: Named in 1 of your videos
Watch it here:
Shows which skill, tool or agent is using up your plan.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A voice dictation app.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Better Creating says it stopped working offline
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
Meeting notes app with no bot in the call. Its MCP lets agents chat with your meeting history.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Meta's everyday agent for online tasks like making bookings.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Official Anthropic skills for working with PDF, Excel, PowerPoint and Word files.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
A server for your audiobooks.
In your videos: Named in 1 of your videos
Watch it here:
Bundled skills for Apple Notes, Apple Reminders, research papers and ASCII video.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Lets you lock ChatGPT chats behind a PIN or biometrics.
In your videos: Named in 2 of your videos
Watch it here:
Meeting notes and summaries inside ChatGPT.
In your videos: Named in 1 of your videos
Watch it here:
Makes slide decks inside ChatGPT, with team comments and export.
Claude dashboards built from BigQuery, Databricks, Snowflake and Salesforce data.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
A Claude Code script that removes ads from a TCL smart TV.
In your videos: Claimed 22,000 stars · Named in 3 of your videos
Watch it here:
A task board that an AI project manager keeps updated.
In your videos: Named in 3 of your videos
Watch it here:
A Mac app that shrinks and converts files on your clipboard. Has an MCP for Claude.
In your videos: Named in 1 of your videos
Watch it here:
Sends files between computers with encryption, and can resume if interrupted.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Lets you chat with Claude Cowork from your phone.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
A self-hosted Matrix chat server and Element app, for talking to Hermes.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A meeting recorder whose transcripts Hermes can pull in.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
On-device dictation and meeting notes. Replaces Granola.
Watch out: Mac only
In your videos: Named in 2 of your videos
Watch it here:
Mac folder rules with conditions and actions, with no AI.
In your videos: Named in 1 of your videos
Watch it here:
Offline meeting notes, and a check for whether content is AI-made, from Google.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Calendar automation for Hermes.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Built-in skills for Drive, Docs, Sheets, Calendar, Gmail and People, with Google sign-in.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Control a dashboard by pinching and holding in front of a camera.
In your videos: Named in 1 of your videos
Watch it here:
Free voice dictation that runs on your Mac.
In your videos: Named in 1 of your videos
Watch it here:
Tracks your posture on a Mac.
In your videos: Named in 1 of your videos
Watch it here:
A floating always-on-top chat box on your Mac that can see your screen.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Teku AI says it only works on the same machine as the agent
In your videos: Named in 4 of your videos
Watch it here:
Lets you message Hermes over iMessage, with a bundled iMessage skill on Mac.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 4 of your videos
Watch it here:
Self-hosted photo backup.
In your videos: Named in 1 of your videos
Watch it here:
Free, open-source alternative to Superhuman email, with an MCP.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Hosted plan $18
In your videos: Named in 1 of your videos
Watch it here:
An open-source media server.
In your videos: Named in 1 of your videos
Watch it here:
Apps for playing music from Jellyfin: Symfonium on Android, plus Finamp and Feishin.
In your videos: Named in 1 of your videos
Watch it here:
A server for ebooks, comics and manga.
In your videos: Named in 1 of your videos
Watch it here:
Finds leftover node_modules and build clutter in old projects.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Issue boards, used as a source for capturing tasks.
In your videos: Named in 1 of your videos
Watch it here:
Browse and unpack nested archives on a Mac.
In your videos: Named in 1 of your videos
Watch it here:
Adds tap gestures to a Mac.
In your videos: Named in 1 of your videos
Watch it here:
Makes bootable USB drives on a Mac.
In your videos: Named in 1 of your videos
Watch it here:
Cleans up Mac disk space, fully uninstalls apps and analyses the disk.
In your videos: Named in 1 of your videos
Watch it here:
Sends a daily summary of your calendar and email to your phone.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
A self-hosted music server.
In your videos: Named in 1 of your videos
Watch it here:
A self-hosted file cloud that works over WebDAV.
In your videos: Named in 1 of your videos
Watch it here:
Handwritten notes on iPad, with transcription.
In your videos: Named in 1 of your videos
Watch it here:
A Linux setup from DHH with an AI agent built in.
In your videos: Named in 3 of your videos
Popularity: 45.7k stars
Watch it here:
Makes an interactive online classroom with slides and AI students you can practise with.
In your videos: Named in 1 of your videos
Watch it here:
Runs your own media server for films and music, with a music app.
In your videos: Named in 1 of your videos
Watch it here:
Remote desktop from your iPad to a Mac.
In your videos: Named in 1 of your videos
Watch it here:
Open-source game-streaming host and client, used for remote desktop over Tailscale.
Watch out: bashbunni rejected Parsec for lacking LAN-only mode
In your videos: Named in 1 of your videos
Watch it here:
A teleprompter on your Mac that shows your script while you read it.
In your videos: Named in 1 of your videos
Watch it here:
Daily commands: one pulls calendar and inbox to set three priorities, another logs your day by voice.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A spreadsheet-style explorer for CSV and other structured data, inside the terminal.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Talk to Hermes out loud and use a wake word to hand it tasks.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Alex Finn says ChatGPT voice mode needs an OpenAI API key, about 5 cents a minute
In your videos: Named in 3 of your videos
Watch it here:
A small toolbox of handy tools that lives in the Mac menu bar.
In your videos: Named in 1 of your videos
Watch it here:
A pack of about 20 to 49 skills for SEO, ads, copywriting and conversion.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Popularity: 56.3k stars
Watch it here:
Open-source alternative to Semrush and Ahrefs with real keyword and ranking data. Pay per use.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 3 of your videos
Watch it here:
A payments connector, also an example of a system watched by cron.
In your videos: Named in 3 of your videos
Watch it here:
Ten finance agents from Anthropic, such as know-your-customer checks.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Anthropic plugins for finance, HR, legal, marketing and operations work.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Finds and fills in details about sales leads.
In your videos: Named in 1 of your videos
Watch it here:
Saves a brand voice and visual rules so output stays consistent.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Runs 11 audit agents at once on paid ads and gives a health score.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: AI Edge cites 9.3K stars · Named in 1 of your videos
Watch it here:
A daily list of quiet or unhappy clients.
In your videos: Named in 1 of your videos
Watch it here:
Turn one piece of content into many formats using scheduled jobs.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Keyword research for Etsy shops.
In your videos: Named in 1 of your videos
Watch it here:
A blog platform, used as a deploy example.
In your videos: Named in 1 of your videos
Watch it here:
Marketing CRM and chat automation, with paid skill packs.
In your videos: Named in 1 of your videos
Watch it here:
SEO data connector for sites that agents build.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A database of business ideas, often paired with Hermes.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A page to publish prototypes and collect usability-test results.
In your videos: Named in 1 of your videos
Watch it here:
An AI acts as your PR manager and finds reporters.
In your videos: Named in 1 of your videos
Watch it here:
Lets you schedule social media posts ahead of time, like Buffer. The name is uncertain.
In your videos: Named in 1 of your videos
Watch it here:
Sends email from code, so an automatic newsletter can go out without you.
In your videos: Named in 1 of your videos
Watch it here:
Checks a website's search ranking setup using 12 smaller skills.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Official Stripe skill that helps Claude with payments. Install just the skill, not the whole plugin.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Drafts and schedules X posts using an API key. A typefully-drafts skill exists.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Posts to many social platforms at once, with a command line and a skill.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
YouTube demand research, used by a YouTube agent.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
A pattern where an LLM turns raw sources into a linked markdown wiki, with ingest, query and lint steps.
Suits: Brainiac, Cerebra. Brainiac maintains the vault and Cerebra indexes it; this is the method behind both.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Lets a model look things up in a vector store, so it can use your private data.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 3 of your videos
Watch it here:
Free Anthropic courses such as Claude 101 and AI Fluency, plus an AI-native SDLC playbook.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
The parts of a local agent: home, channel, brain, memory, skills, heartbeat and eyes.
An agent needs a clear goal, skills, tools and context to work on its own.
In your videos: Named in 1 of your videos
Watch it here:
How agents talk to each other.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Free courses from Google, deeplearning.ai, Elements of AI, OpenAI, Hugging Face, LangChain Academy and Kaggle.
In your videos: Named in 1 of your videos
Watch it here:
A book list on AI, plus Karpathy's Zero to Hero videos.
In your videos: Named in 1 of your videos
Watch it here:
Choosing exactly what goes into a model's context.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
One measurable sentence saying what done looks like for each agent task.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A free YouTube playlist on the minimum maths you need for AI.
In your videos: Named in 1 of your videos
Watch it here:
A catalogue of agent loops that measure results and repeat.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Four layers, Memory, Agent, Pulse and Screen, for turning Claude Code into an operating system.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
The standard way agents and servers share tools, resources and prompts. It is set up in mcp.json.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A tiny model that can look at webcam frames and describe what it sees.
Animated avatars for your bots in Hermes bot mode.
In your videos: Named in 1 of your videos
Watch it here:
Usable RAM is 75% of RAM. Divide by 0.6 to get billions of parameters that fit.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Pick tasks that are repetitive, rules-based and give time back.
In your videos: Named in 1 of your videos
Watch it here:
The six parts of an agent: model, tools, memory, audio, guardrails and orchestration.
In your videos: Named in 1 of your videos
Watch it here:
A file-sharing method chosen over SSHFS, NFS and SMB, run with a Docker WebDAV server.
In your videos: Named in 1 of your videos
Watch it here:
Spots objects in camera images on a phone using the Ultralytics app.
All 603 items in one alphabetical list. Each one shows which chapter it lives in.
Android Debug Bridge, used to connect an Android phone to Hermes.
In your videos: Named in 1 of your videos
Watch it here:
About 24 senior-engineer skills with commands to spec, plan, build, test, review and ship.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Popularity: 110.0k stars
Watch it here:
Lets a cheap model ask a stronger one, like Opus, when it gets stuck.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
279 ready-made role-based agents across company divisions.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: AI Edge cites 152k stars · Named in 1 of your videos
Watch it here:
Each agent writes down what it did, or adds it to today's daily note, before replying.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 3 of your videos
Watch it here:
The parts of a local agent: home, channel, brain, memory, skills, heartbeat and eyes.
An agent needs a clear goal, skills, tools and context to work on its own.
In your videos: Named in 1 of your videos
Watch it here:
Export what your AI system knows about you as markdown, to set up a new agent.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
A live building picture showing which agents are working or idle.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
How agents talk to each other.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Compare an agent's output with a quality bar or SOP kept in skills.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Dan Martell's steps for a first agent: Aim, Give identity, Equip, Narrow, Trust, with staged trust.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A folder of business markdown docs that SOUL.md tells the agent to read.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Paid dashboard that watches Claude, Codex, Hermes and Antigravity agents, with steer and stop buttons.
Watch out: Sold inside a paid community
In your videos: Named in 3 of your videos
Watch it here:
Lets agents read sites like X, Reddit, YouTube and GitHub without paying for API fees.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Teku AI cites about 81k stars and says it gets past 403 bot blocks · Named in 4 of your videos
Popularity: 101.3k stars
Watch it here:
A site that explains what agent skills are.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
One shared instruction file for agents. Claude Code now reads it, and it can sit in any folder.
Suits: Forge, Brainiac. Forge builds and runs agent operations, where this belongs; Brainiac could also use it.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Tell the agent to write everything it does into daily logs.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Free courses from Google, deeplearning.ai, Elements of AI, OpenAI, Hugging Face, LangChain Academy and Kaggle.
In your videos: Named in 1 of your videos
Watch it here:
A book list on AI, plus Karpathy's Zero to Hero videos.
In your videos: Named in 1 of your videos
Watch it here:
A database for research sections and citations inside n8n.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
Open-source code review, with a VS Code extension.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Official skills for building Android apps with coding agents.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Free Anthropic courses such as Claude 101 and AI Fluency, plus an AI-native SDLC playbook.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Official Anthropic skills for working with PDF, Excel, PowerPoint and Word files.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Ten finance agents from Anthropic, such as know-your-customer checks.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Anthropic plugins for finance, HR, legal, marketing and operations work.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Google's code editor built around agents, often compared with Hermes.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
Local workspace where you upload documents and run agent workflows.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Dan Martell's second-brain product.
In your videos: Named in 1 of your videos
Watch it here:
Thousands of ready-made scrapers that agents can use.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 13.2k stars
Watch it here:
Finds and fills in details about sales leads.
In your videos: Named in 1 of your videos
Watch it here:
Makes the agent ask you first before it spends money, sends things or deletes anything.
Suits: Sentinel, Forge. Sentinel enforces safety and Forge runs ops; gates match Chris's no-surprise rules.
In your videos: Named in 2 of your videos
Watch it here:
Turns plain English into editable HTML diagrams of how a system or workflow fits together.
Suits: Cyborg, Mantis. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches; Mantis could also use it.
In your videos: AI Edge cites 64.4K stars · Named in 2 of your videos
Watch it here:
A browser playground and leaderboard for comparing models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Runs 100 answer attempts in a bracket and keeps the best.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Free Adobe-style apps built in Rust, with a built-in MCP so agents can run them.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Mostly early alpha
In your videos: Named in 4 of your videos
Watch it here:
A desktop tab that collects every link, image and file you sent the agent.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
A server for your audiobooks.
In your videos: Named in 1 of your videos
Watch it here:
Karpathy's loop that runs experiments overnight while you sleep.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A curated list of Claude skills.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Popularity: 15.5k stars
Watch it here:
Runs a second task in the background at the same time.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Optional hand-gesture 3D screen for your files, part of the Jarvis project.
In your videos: Named in 1 of your videos
Watch it here:
Agent memory kept as plain markdown files with wiki links. It replaces Notion.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
Watch out: Hosted plan costs $15 a month
In your videos: Named in 1 of your videos
Popularity: 3.6k stars
Watch it here:
Pauses risky deletes, lists the affected files and waits for you to press Proceed or Cancel.
Suits: Sentinel. Sentinel would like a hard stop on deletes; it fits Chris's archive-not-delete rule.
In your videos: Named in 1 of your videos
Watch it here:
Lets an agent control Blender to make 3D models and games.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Popularity: 32.3k stars
Watch it here:
A diagramming tool to plan an agent before building it.
In your videos: Named in 1 of your videos
Watch it here:
Turns a prompt into a web app, with MCP connectors.
In your videos: Named in 3 of your videos
Watch it here:
Checks a new skill for hidden bad instructions before you install it.
Suits: Sentinel. Sentinel already security-vets skills; Bouncer automates that first check.
In your videos: Named in 1 of your videos
Watch it here:
Turns a project into a 20-second coded launch video; free on GitHub.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Folders of markdown notes with README files that agents search, read and write back to.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Saves a brand voice and visual rules so output stays consistent.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Hosted browser plugin so Hermes can automate the web.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Connects an agent to your real Chrome browser and fixes itself when steps fail.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 18.3k stars
Watch it here:
Browser automation plugin for Hermes.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 119.1k stars
Watch it here:
Bundled skills for Apple Notes, Apple Reminders, research papers and ASCII video.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
A bundled skill that gets a video transcript for summaries or scheduled quizzes.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A usage number shown in the corner of the Claude Code screen.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Block's team workspace that feels like Slack, where agents are members. You can host it yourself.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claimed 35,000 stars · Named in 3 of your videos
Watch it here:
Warns you about the cost before you send a message once the prompt cache has gone cold.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Record yourself doing a job, then turn it into an agent playbook.
In your videos: Named in 1 of your videos
Watch it here:
A design editor, also available as an MCP in Hermes.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Canva's image model, with a Dream Lab tab and an API.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Makes social media graphics and saves them as PNG or PDF.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
An hourly routine gathers Slack, meetings and email into an inbox, then a librarian step files it.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
An ultra-short answer mode that claims about 65% fewer output tokens.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Nate Herk's test found only about 8.5% savings
In your videos: Teku AI cites about 100k stars · Named in 3 of your videos
Popularity: 114.0k stars
Watch it here:
Plugins linking ChatGPT or Codex to Gmail, Notion, Slack, Drive, Calendar, Supabase and Netlify.
In your videos: Named in 2 of your videos
Watch it here:
Chrome extension that lets ChatGPT or Codex control your browser and computer.
In your videos: Named in 3 of your videos
Watch it here:
Always-on OpenAI agent in the ChatGPT desktop app that can watch over Codex threads.
In your videos: Alex Finn calls it incredible · Named in 5 of your videos
Watch it here:
and 1 more videos
Lets you lock ChatGPT chats behind a PIN or biometrics.
In your videos: Named in 2 of your videos
Watch it here:
Meeting notes and summaries inside ChatGPT.
In your videos: Named in 1 of your videos
Watch it here:
Makes slide decks inside ChatGPT, with team comments and export.
A docs-like space in ChatGPT where Dots logs milestones and bugs.
ChatGPT's agent mode with a computer, files and browser. It runs scheduled cloud workflows.
In your videos: Alex Finn says it has the best computer use · Named in 3 of your videos
Watch it here:
Free monitoring for Linux, Docker, Proxmox and websites, with checks for agent health.
Suits: Cyborg, Sentinel. Cyborg runs the LAN telemetry dashboard and Sentinel watches health; fits Chris's planned agent dashboard.
In your videos: NetworkChuck uses it to monitor his Hermes agents (crons, gateway, model login) · Named in 1 of your videos
Watch it here:
Runs 11 audit agents at once on paid ads and gives a health score.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: AI Edge cites 9.3K stars · Named in 1 of your videos
Watch it here:
Anthropic's coding agent for the terminal, with skills, plugins, hooks, MCP and helper agents.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Pav Rusovs chose it over Hermes, citing Anthropic rules on subscriptions in third-party harnesses · Named in 39 of your videos
Popularity: 153.3k stars
Watch it here:
and 35 more videos
Built-in skills that pass coding jobs to Claude Code or Codex, only when you ask.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Runs Claude Code tasks on Anthropic's machines, and pulls the session back when you want.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Keeps looping until a condition is met, and needs proof shown in the chat.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Small scripts that run on events in Claude Code to check work or add context.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 3 of your videos
Watch it here:
A built-in command that reviews your recent Claude Code sessions.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Add-ons that change how Claude Code looks and behaves.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Mods are temporary until you ask Claude to save one as a plugin
In your videos: Named in 1 of your videos
Watch it here:
Beta: one main conversation that starts cloud threads.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A marketplace of skills, agents, MCPs and plugins for Claude Code.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Popularity: 30.8k stars
Watch it here:
A no-code agent inside Claude Desktop with plugins, connectors, scheduled tasks and computer control.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Tina Huang calls it the safer agent she lets read her email · Named in 6 of your videos
Watch it here:
and 2 more videos
Claude's design canvas for mocking up apps and screens with colour options.
Suits: Mantis. Mantis does visual work, and Chris wants drafts made in Claude Design.
In your videos: Named in 2 of your videos
Watch it here:
An always-on Anthropic agent on Opus 5.5 with its own cloud computer, channels and to-do list.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Waitlist at the time
In your videos: Named in 3 of your videos
Watch it here:
Anthropic's strongest tier, named as an advisor model above Opus.
Watch out: Jason Lee hit usage limits with it
In your videos: Named in 12 of your videos
Watch it here:
and 8 more videos
Anthropic's fast, cheap model. Good for small coding helper agents where cost matters.
Suits: Alfred, Brainiac. All seven agents run on Haiku today.
In your videos: Sharbel suggests Haiku sub-agents to protect the main cache · Named in 8 of your videos
Watch it here:
and 4 more videos
Claude dashboards built from BigQuery, Databricks, Snowflake and Salesforce data.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
A file of instructions Claude reads. Keep it a short map to other files, and check it now and then.
Suits: Forge, Brainiac. Forge builds and runs agent operations, where this belongs; Brainiac could also use it.
Watch out: Mason Fitzgerald calls it a soft instruction that is not enforced
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
Keeps a lasting memory for Claude Code so it remembers things between sessions.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
Watch out: Nate Herk says skip it: it sends data by default
In your videos: Named in 1 of your videos
Watch it here:
Turns a prompt into an editable animation that you tweak with comments.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Beta for Team and Enterprise plans only
In your videos: Named in 2 of your videos
Watch it here:
Anthropic's top model line. Opus 5.5 is the one Anthropic suggests as your default.
Suits: Forge. Forge could use it as a strong orchestrator; the playlist-1 doc notes Chris's agents all run Haiku today.
Watch out: Alex Finn says it costs him about 40 USD a day; Moe Lueker says avoid it for routine tasks
In your videos: Named in 24 of your videos
Watch it here:
and 20 more videos
Anthropic's official repository of Claude Code plugins.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 44.4k stars
Watch it here:
Cloud jobs that start on a schedule, an API link with a token, or a GitHub event.
Suits: Brainiac, Cyborg. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cyborg could also use it.
In your videos: Greg Isenberg's guest calls it Cowork scheduled tasks on steroids · Named in 2 of your videos
Watch it here:
Anthropic's multi-agent review that checks your code for security problems.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 2 of your videos
Watch it here:
Changes how Claude Code shows its tool calls, with seven colour themes to pick from.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Anthropic's middle model, called the consensus best choice to power Hermes.
Suits: Forge, Cerebra. Forge builds and runs agent operations, where this belongs; Cerebra could also use it.
In your videos: Dan Martell: Sonnet to run agents, Opus to build them · Named in 13 of your videos
Watch it here:
and 9 more videos
Runs several coding agent sessions side by side in the terminal.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Lets an agent watch a video from a link or file.
Suits: Cerebra, Mantis. Cerebra does web and source-graded research and vault indexing, which this feeds; Mantis could also use it.
In your videos: Named in 1 of your videos
Watch it here:
A Claude Code script that removes ads from a TCL smart TV.
In your videos: Claimed 22,000 stars · Named in 3 of your videos
Watch it here:
Resets the Claude Code context for free.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Builds a command-line tool for any desktop app so agents can drive it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A task board that an AI project manager keeps updated.
In your videos: Named in 3 of your videos
Watch it here:
A daily list of quiet or unhappy clients.
In your videos: Named in 1 of your videos
Watch it here:
A Mac app that shrinks and converts files on your clipboard. Has an MCP for Claude.
In your videos: Named in 1 of your videos
Watch it here:
Manages your domains and subdomains. DDNS updates them when your public IP changes.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Turns a coding agent into a security auditor.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Makes a local model or app reachable from the internet.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
A server control panel template available on Hostinger.
In your videos: Named in 1 of your videos
Watch it here:
Turns your code into a lasting knowledge graph so the agent uses fewer tokens exploring.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claims 99% (Sharbel) or 120x (Teku AI) fewer tokens; about 45k stars · Named in 4 of your videos
Popularity: 47.2k stars
Watch it here:
Shows where your Claude tokens go and flags MCP servers you never use.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Meta's coding models, including small versions you can run locally.
In your videos: Named in 2 of your videos
Watch it here:
OpenAI's coding agent, in the terminal or the ChatGPT desktop app.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Samin Yasar uses Codex as the doctor that fixes Hermes · Named in 35 of your videos
Popularity: 128.1k stars
Watch it here:
and 31 more videos
Lets Codex read the iMessage database stored on your Mac.
In your videos: Named in 1 of your videos
Watch it here:
Connects agents to over 1,000 apps like Gmail, Slack and Notion through sign-in, and has a command line.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 5 of your videos
Popularity: 30.7k stars
Watch it here:
and 1 more videos
A settings panel and a manual fork that shares the cache.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Safe, sealed-off workspaces from Dagger where an agent can work without touching your real machine.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Popularity: 4.0k stars
Watch it here:
Turn one piece of content into many formats using scheduled jobs.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Shows what is filling the context window.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Feeds the agent current documentation for code libraries.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 63.7k stars
Watch it here:
Checks Hermes for clutter in its files, memory and tools, and sets items aside instead of deleting.
Suits: Forge, Sentinel. Forge runs agent ops and Sentinel reviews skills, so both benefit from a bloat audit that archives instead of deleting.
In your videos: Featured as the headline skill; the playlist-1 doc made it a top pick · Named in 1 of your videos
Watch it here:
Choosing exactly what goes into a model's context.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Saves space in Claude Code's working memory so long sessions last longer.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Nate Herk says skip it: usage-limit bug
In your videos: Named in 1 of your videos
Watch it here:
Community Obsidian plugin for chatting with AI about your vault.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Agent memory system and code analyzer that works with Graphify.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Shows what the current session has cost.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An open-source web scraper, the second choice after the main one.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Sends files between computers with encryption, and can resume if interrupted.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Lets scheduled jobs remember earlier runs so they do not repeat work.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A scheduled-job option that runs only a script, saving tokens by skipping the AI.
Suits: Brainiac. Brainiac's scheduled vault jobs could run as plain scripts.
In your videos: Named in 1 of your videos
Watch it here:
An AI code editor with cloud agents. It can also use MCP tools.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
A custom Claude Code command that connects to your server to edit and manage a Hermes agent.
Suits: Forge. Forge could manage the Hermes agents from Claude Code the same way.
In your videos: Named in 1 of your videos
Watch it here:
Scheduled scripts write numbers into daily-note headers so your notes work like a table you can query.
Suits: Brainiac, Cyborg. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cyborg could also use it.
In your videos: Named in 2 of your videos
Watch it here:
A scheduled job scans Reddit and X for business ideas and builds a dashboard.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Lets Hermes control DaVinci Resolve Studio live for video editing.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Multi-agent research inside Claude Code that cross-checks its findings.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A multi-step research process that uses several models at once.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
A very large open model made for coding.
An open model built to reason step by step.
Cheap open models that use tools well. Popular choice to power Hermes.
Suits: Alfred, Cerebra. Alfred, the general assistant, could use it in everyday chats; Cerebra could also use it.
In your videos: Metics Media picks V4 Flash as a cheap main model · Named in 16 of your videos
Watch it here:
and 12 more videos
One measurable sentence saying what done looks like for each agent task.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Strips a web page down to clean readable text, which saves on tokens.
Suits: Cerebra, Brainiac. Cerebra fetches pages for research and Brainiac clips into the vault; both save tokens.
In your videos: Named in 4 of your videos
Popularity: 9.4k stars
Watch it here:
About 18 skills that send tasks to coding agents you have installed, like Claude Code, Codex and Cursor.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A file view with a side-by-side site preview where you click parts to leave comments.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An autonomous coding agent, listed inside Kylon.
In your videos: Named in 1 of your videos
Watch it here:
Turns one sentence into a designed diagram, in about 45 types.
Suits: Mantis, Cyborg. Mantis does image prompts and creative work, which this produces; Cyborg could also use it.
In your videos: Named in 2 of your videos
Watch it here:
A chat app where your Hermes agents live, with one channel per project.
Suits: Alfred. Chris's agents already talk in Discord channels.
In your videos: Named in 12 of your videos
Watch it here:
and 8 more videos
Lets you chat with Claude Cowork from your phone.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Checks whether a DNS change has spread across public resolvers.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Helps Claude write documents together with you.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Runs several apps and agents separately on one machine. Docker Compose handles whole stacks.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Runs agent code and browsing inside Docker so it stays walled off from the rest of your server.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Agents save long reports in a documents area, not chat, which saves tokens.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A self-hosted panel to deploy apps, databases and stacks, with GitHub auto-deploy and previews.
Suits: Forge. Forge runs deploys; Dokploy is an alternative to Chris's Coolify.
In your videos: Named in 2 of your videos
Watch it here:
One safe place to store your API tokens. It syncs to Vercel and has a skill.
Suits: Sentinel. Sentinel would keep secrets out of chats and Discord.
In your videos: Named in 1 of your videos
Watch it here:
The agent writes email replies in your voice but never sends them on its own.
Suits: Alfred, Sentinel. Alfred, the general assistant, could use it in everyday chats; Sentinel could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Fast local image generation on Apple silicon Macs and iPhone.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Runs dozens of agents at once from one script. Set the size small in the settings.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A big bundle of 292 skills that guide Claude through plan, test, build, review and check steps.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A self-hosted Matrix chat server and Element app, for talking to Hermes.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Cloud voices, music and lip-sync (Chimera), available through an API.
Suits: Alfred, Mantis. Alfred, the general assistant, could use it in everyday chats; Mantis could also use it.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Keyword research for Etsy shops.
In your videos: Named in 1 of your videos
Watch it here:
Official Expo skill that helps Claude build mobile apps.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Backup models and accounts so your agent keeps working when one provider fails.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Python kit for building your own MCP servers and clients.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 24.3k stars
Watch it here:
Adds a file panel on the right that shows which files Claude touched and which are saved.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Each workflow folder holds lessons learned, which the agent reads before every run.
Suits: Brainiac, Forge. Brainiac files and maintains the vault and runs cron jobs, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Turns websites into clean markdown. It can scrape, crawl and get past bot detection.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: The Next New Thing offers Scrapling as a free replacement
In your videos: Named in 4 of your videos
Popularity: 194.9k stars
Watch it here:
A meeting recorder whose transcripts Hermes can pull in.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Text-to-speech voices, including one steered by bracketed directions.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
On-device dictation and meeting notes. Replaces Granola.
Watch out: Mac only
In your videos: Named in 2 of your videos
Watch it here:
Black Forest Labs image models for making pictures from text.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Watch it here:
Mac folder rules with conditions and actions, with no AI.
In your videos: Named in 1 of your videos
Watch it here:
A local helper that cuts token costs by sending simple calls to a cheaper route.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Weak until trained
In your videos: Named in 2 of your videos
Watch it here:
Chains about 635 free AI providers together. Comes as a desktop and Android app.
Watch out: Nuno Tavares says skip: the free models are weak
In your videos: Named in 2 of your videos
Watch it here:
Anthropic skill that avoids generic fonts and forces a clear design direction.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Animates 3D characters with loose clothing or animal bodies.
In your videos: Named in 1 of your videos
Watch it here:
Google's model family, used as cheap writers and agent drivers.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: Samin Yasar says skip it
In your videos: Tina Huang likes Gemini Flash for writing into the vault · Named in 6 of your videos
Watch it here:
and 2 more videos
Google's top model, priced at $2 in and $10 out per million tokens for a limited time.
In your videos: Named in 1 of your videos
Watch it here:
Small AI models that run on the device itself, not in the cloud.
In your videos: Named in 1 of your videos
Watch it here:
A notebook that answers from your own sources, and makes audio, video, slides, mind maps and reports.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Google's open model for phones and local use, including uncensored versions.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
A blog platform, used as a deploy example.
In your videos: Named in 1 of your videos
Watch it here:
Lets an agent operate apps on your Mac.
In your videos: Named in 1 of your videos
Watch it here:
A terminal program that Alex Finn uses.
In your videos: Named in 1 of your videos
Watch it here:
AI coding helper, listed among coding tools.
In your videos: Named in 1 of your videos
Watch it here:
Plays your Git history back as an animated replay.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Saves versions of your agent or vault, with .gitignore, and backs it up daily to GitHub.
Suits: Cyborg. Cyborg already publishes the vault to GitHub.
In your videos: Named in 3 of your videos
Watch it here:
Separate git worktrees let several agents work on one repo at the same time.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Company-wide search plus an agent layer.
In your videos: Named in 1 of your videos
Watch it here:
An open model from Z.AI. It is the cheapest option Alex Finn names.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 9 of your videos
Watch it here:
and 5 more videos
A browser console showing live plane, ship and satellite data.
In your videos: Named in 1 of your videos
Watch it here:
Marketing CRM and chat automation, with paid skill packs.
In your videos: Named in 1 of your videos
Watch it here:
Guidelines from JetBrains that make Claude Code or Codex write up-to-date Go code.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Google's kit for building agents. It supports open models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Offline meeting notes, and a check for whether content is AI-made, from Google.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A phone app to download and run Gemma 4 locally.
Calendar automation for Hermes.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Free notebooks with a T4 GPU, used to run or fine-tune models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Google's game builder.
In your videos: Named in 1 of your videos
Watch it here:
SEO data connector for sites that agents build.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Lets an agent use Drive, Gmail, Calendar, Docs, Sheets and Slides, with over 40 skills.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Popularity: 30.4k stars
Watch it here:
Built-in skills for Drive, Docs, Sheets, Calendar, Gmail and People, with Google sign-in.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
A small OpenAI model used to test n8n agents.
In your videos: Named in 1 of your videos
Watch it here:
ChatGPT-plan models you can use in Hermes by logging in, with no API cost.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Alex Finn says 5.5 is the first usable cheap option via a ChatGPT subscription · Named in 11 of your videos
Watch it here:
and 7 more videos
OpenAI's top model, strong at using computers and coding. It powers ChatGPT Dots.
In your videos: Alex Finn says only Opus 5.5 is ahead of it · Named in 13 of your videos
Watch it here:
and 9 more videos
OpenAI's image maker, also usable through a Hermes plugin and a Codex skill.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
OpenAI's light, cheap model tier. Good value for everyday agent work.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 4 of your videos
Watch it here:
OpenAI's open-weight model. The 120B size needs large local hardware.
In your videos: Named in 2 of your videos
Watch it here:
An agent that researches a topic on its own and writes a report.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 29.5k stars
Watch it here:
OpenAI's top tier, used for heavy monthly reviews.
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
OpenAI's middle-tier model.
In your videos: Named in 2 of your videos
Watch it here:
Meeting notes app with no bot in the call. Its MCP lets agents chat with your meeting history.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
A tool that builds a knowledge graph. Named as Cortex Suite's partner.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Builds a regex pattern for you from example strings.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
xAI's model, linked by X login. It can read and search X.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 8 of your videos
Watch it here:
and 4 more videos
xAI's agent app. Each bot gets its own cloud computer, routines and reports.
In your videos: Named in 10 of your videos
Watch it here:
and 6 more videos
xAI's image maker, available through an API and used for thumbnails.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
A voice agent model with a free sandbox to try it.
In your videos: Named in 1 of your videos
Watch it here:
A hosted playground and fast API for running AI models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Adds footnote-style sources to research results so you can see where facts came from.
Suits: Cerebra. Cerebra's source-graded research needs citations.
In your videos: Named in 1 of your videos
Watch it here:
A shared chat for several bots, with a turn limit before you must step in.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Garry Tan's chain of role skills for planning, reviewing and shipping software.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: AI Edge cites 133k stars · Named in 3 of your videos
Popularity: 137.3k stars
Watch it here:
Control a dashboard by pinching and holding in front of a camera.
In your videos: Named in 1 of your videos
Watch it here:
Writes a short handoff note of a session so another agent can carry on.
Suits: Alfred, Forge. Alfred and Forge both hand work between agents and surfaces.
In your videos: Named in 3 of your videos
Watch it here:
Ask for a handoff file at around 200 to 300k tokens, then restart in a fresh chat.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Free voice dictation that runs on your Mac.
In your videos: Named in 1 of your videos
Watch it here:
Tracks your posture on a Mac.
In your videos: Named in 1 of your videos
Watch it here:
A skill that hosts and deploys websites your agent has made.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Nous Research's open model family for running locally.
In your videos: Named in 1 of your videos
Watch it here:
The main program that runs your always-on agents. It has memory, skills, scheduled jobs and over 20 chat channels.
Suits: Alfred, Brainiac, Cerebra. It is the runtime all seven of Chris's agents already use.
In your videos: Alex Finn calls it the greatest AI tool ever made · Named in 56 of your videos
Popularity: 253.6k stars
Watch it here:
and 52 more videos
Runs several specialist bots, each with its own personality, skills, memory and model, that can mention each other.
Suits: Forge. Forge runs the agent team; bot mode is how the seven agents could share rooms.
In your videos: Sharbel calls it a cheat code; Wanderloots suggests keeping 3 to 5 bots warm and a 10 to 30 minute idle timeout · Named in 3 of your videos
Watch it here:
Opens the Hermes settings to edit, for example to limit admin commands to one person.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Runs a task on a timer by itself and sends the result to a chat channel.
Suits: Brainiac. Brainiac runs Chris's cron work.
In your videos: Alex Finn says to prune and pause old cron jobs to improve speed and cost · Named in 13 of your videos
Watch it here:
and 9 more videos
A desktop app for using Hermes: chats, models, profiles, scheduled jobs, skills and pinned sessions.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 9 of your videos
Watch it here:
and 5 more videos
A health-check command that reports whether your Hermes install is working.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Reviews recent activity across your agent tools and gathers the lessons learned.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Restarts the gateway so your config and channel changes take effect.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Ways to connect the Hermes app to an agent: local, Hermes Cloud, remote gateway or SSH.
Suits: Forge. Forge runs the Coolify-hosted agents the app would connect to.
In your videos: Teku AI uses the remote gateway to reach a VPS · Named in 1 of your videos
Watch it here:
Keeps the agent working until your stated goal is done, with a limit on turns.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Jack Roberts says a goal has a 20-turn budget by default
In your videos: Sharbel says to have Hermes write the goal prompt first (meta-prompting) · Named in 5 of your videos
Watch it here:
and 1 more videos
A task board where work moves from new to done and gets shared out between helper agents.
Suits: Forge, Brainiac. Forge runs builds and ops; Brainiac could own filing tasks on the board.
In your videos: Named in 10 of your videos
Watch it here:
and 6 more videos
Turns a PDF, link or YouTube video into a new saved skill for your agent.
Suits: Brainiac, Cerebra. Brainiac and Cerebra could turn sources into reusable skills.
In your videos: Named in 2 of your videos
Watch it here:
Adds one outside memory service, such as Honcho, Mem0 or Supermemory, to Hermes.
Suits: Brainiac, Cerebra. Brainiac owns memory and Cerebra indexes the vault.
Watch out: Sharbel says only one external provider can be active at a time
In your videos: Named in 2 of your videos
Watch it here:
A searchable wiki of decisions, logs and research that your agent keeps up to date so it stays consistent.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 4 of your videos
Watch it here:
Jack Roberts' set-up: a team of personas over Hermes, each with its own job, prompt and model.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A screen to browse official and community Hermes plugins with badges and reviews.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 3 of your videos
Watch it here:
Separate Hermes setups, each with its own keys, memory, scheduled jobs, skills and personality.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Alex Finn recommends at least two profiles so one can fix the other · Named in 6 of your videos
Watch it here:
and 2 more videos
Terminal commands to set up your AI providers and switch the model in use.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
Turns tools such as browser automation or computer use on or off for each platform.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 2 of your videos
Watch it here:
A web page to see and manage your Hermes sessions, models, jobs, skills, keys, channels and logs.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Sharbel's mission-control dashboard for agents, hosted on Vercel and gated by Google sign-in.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
A platform for deploying apps.
In your videos: Named in 1 of your videos
Watch it here:
A low-cost server provider, about 5 to 6 dollars a month.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Hetzner's command line plus an API token, so Claude Code can create and manage servers.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
AI avatar video, combined with Remotion.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Makes images and video with major models through an MCP connector and command line, including face-consistent Soul.
Suits: Mantis. Mantis makes images and could generate them directly.
In your videos: Tristen O'Brien adds it to Claude as a custom connector · Named in 6 of your videos
Watch it here:
and 2 more videos
Long-term agent memory that can retain, recall and reflect. A Hermes memory option that runs in Docker.
Suits: Brainiac, Cerebra. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cerebra could also use it.
Watch out: Playlist-1 doc notes it overlaps Chris's Qdrant plan
In your videos: Named in 4 of your videos
Popularity: 44.6k stars
Watch it here:
Coding agent repo that reviews its own past sessions. It may be the same Hindsight.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Memory service shared across agents, with dreaming. Check it with the command hermes honcho status.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 4 of your videos
Popularity: 7.8k stars
Watch it here:
A rented server with one-click Hermes, Dokploy and n8n templates and the Kodee assistant.
Suits: Forge. Forge runs hosting, but Chris already hosts on Coolify, so this is only a comparison.
Watch out: Lead Gen Jay calls Hostinger servers weak
In your videos: Named in 12 of your videos
Watch it here:
and 8 more videos
The same repo on every machine lists your projects, so threads can start remotely.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 1 of your videos
Watch it here:
A floating always-on-top chat box on your Mac that can see your screen.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Teku AI says it only works on the same machine as the agent
In your videos: Named in 4 of your videos
Watch it here:
A hub for open models and Spaces demos you can try.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
A rule that agents may not edit the human side of your vault. The machine side is theirs.
Suits: Brainiac, Sentinel. Brainiac and Sentinel already guard off-limits folders like the archive.
In your videos: Named in 1 of your videos
Watch it here:
Rewrites text to remove the usual signs that an AI wrote it.
Suits: Alfred, Mantis. Alfred and Mantis write text for Chris; it is already built into Hermes.
Watch out: Teku AI says it is token heavy; Sharbel notes the skill file is 29.6 KB
In your videos: Playlist-1 doc notes it is already built into Hermes · Named in 4 of your videos
Popularity: 60.1k stars
Watch it here:
Double-click commands that draft with a local model into a review folder before anything is applied.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Tencent's open model, free on some providers.
In your videos: Named in 3 of your videos
Watch it here:
A workspace where a team of agents works inside Slack. Sign in with a ChatGPT plan.
In your videos: Named in 3 of your videos
Watch it here:
The agent writes an HTML page that turns into an MP4 video.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Needs Node 22
In your videos: Named in 7 of your videos
Popularity: 71.8k stars
Watch it here:
and 3 more videos
Copies the format of a viral video.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Nuno Tavares says skip: needs paid services
In your videos: Named in 4 of your videos
Watch it here:
A free YouTube playlist on the minimum maths you need for AI.
In your videos: Named in 1 of your videos
Watch it here:
A database of business ideas, often paired with Hermes.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Makes editable SVG images, which are sharp at any size.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Lets you message Hermes over iMessage, with a bundled iMessage skill on Mac.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 4 of your videos
Watch it here:
Self-hosted photo backup.
In your videos: Named in 1 of your videos
Watch it here:
Gives Claude 24 design commands, like critique and delight, to improve how web pages look.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Free, open-source alternative to Superhuman email, with an MCP.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Hosted plan $18
In your videos: Named in 1 of your videos
Watch it here:
Open-source Claude Code assistant with voice, a visualizer, a persona and memory kept in an Obsidian vault.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
An open-source media server.
In your videos: Named in 1 of your videos
Watch it here:
Apps for playing music from Jellyfin: Symfonium on Android, plus Finamp and Feishin.
In your videos: Named in 1 of your videos
Watch it here:
A cheap, fast model for making decisions and scoring. The name may be wrong.
In your videos: Named in 3 of your videos
Watch it here:
A pattern where an LLM turns raw sources into a linked markdown wiki, with ingest, query and lint steps.
Suits: Brainiac, Cerebra. Brainiac maintains the vault and Cerebra indexes it; this is the method behind both.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
A server for ebooks, comics and manga.
In your videos: Named in 1 of your videos
Watch it here:
Never paste keys into chat. Set them as environment variables in the terminal and give least access.
Suits: Sentinel. Sentinel enforces Chris's no-secrets rule.
In your videos: Named in 1 of your videos
Watch it here:
Moonshot's open model, used as a cheap option to power Hermes.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 11 of your videos
Watch it here:
and 7 more videos
Local text-to-speech voice. Slower than Piper.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Finds leftover node_modules and build clutter in old projects.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An image model that lets you edit inside the picture. In public beta.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Slack-style workspace where people and Claude Code or Codex agents share channels and automations.
In your videos: Matt Wolfe found it impressive; Nate Herk notes it was sponsored · Named in 2 of your videos
Watch it here:
A page to publish prototypes and collect usability-test results.
In your videos: Named in 1 of your videos
Watch it here:
A framework for building agents and apps on language models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Visual builder where you draw an agent pipeline instead of writing code.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Open framework for building agents as connected graphs of steps.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Researches a topic from the last 30 days across Reddit, X, YouTube, GitHub and the web.
Suits: Cerebra. Cerebra does web research and this gives her fresh multi-platform sentiment.
In your videos: AI Edge notes 140 contributors; Futurepedia lists it first · Named in 3 of your videos
Popularity: 65.2k stars
Watch it here:
Gives you free SSL certificates for your sites.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Agent framework with memory built in.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Issue boards, used as a source for capturing tasks.
In your videos: Named in 1 of your videos
Watch it here:
Meta's open models, including large and uncensored versions.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Runs local models directly and faster. Ollama is built on top of it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A framework for building apps and agents that use your data with language models.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Karpathy's project where models with different personas debate to improve a decision.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 25.6k stars
Watch it here:
Checks your computer's hardware and suggests local AI models that will fit.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Hermes skill where a librarian bot files reports into your Obsidian wiki and asks you to approve reviews.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 2 of your videos
Watch it here:
A desktop app to download local AI models and chat with them offline.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Lower GPU offload and context length if it fails to load a model
In your videos: Named in 3 of your videos
Watch it here:
A self-built MCP that shows agents only the API routes you allow, such as Redmine.
Suits: Sentinel, Forge. Sentinel would want least-privilege API access for every agent.
In your videos: Named in 1 of your videos
Watch it here:
Finds your computer hardware and picks and loads the best local model for it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A settings file that blocks sending, sharing and deleting on the server. Email stays draft only.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A catalogue of agent loops that measure results and repeat.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An app builder for coding by vibes.
In your videos: Named in 2 of your videos
Watch it here:
Browse and unpack nested archives on a Mac.
In your videos: Named in 1 of your videos
Watch it here:
Adds tap gestures to a Mac.
In your videos: Named in 1 of your videos
Watch it here:
Makes bootable USB drives on a Mac.
In your videos: Named in 1 of your videos
Watch it here:
Uses a cheap model for background jobs while the main model handles your chat.
Suits: Forge. Forge sets models; this saves OpenRouter credit for all seven agents.
In your videos: Named in 1 of your videos
Watch it here:
Security review skills, said to be by Google, that can run on a schedule.
Suits: Sentinel. Sentinel does security review; note the name clashes with Chris's Mantis agent.
In your videos: Named in 1 of your videos
Watch it here:
A web agent that does tasks on its own.
In your videos: Named in 1 of your videos
Watch it here:
Four layers, Memory, Agent, Pulse and Screen, for turning Claude Code into an operating system.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A prompt structure: Mission, Ask, Parameters, Shape.
In your videos: Named in 1 of your videos
Watch it here:
A pack of about 20 to 49 skills for SEO, ads, copywriting and conversion.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Popularity: 56.3k stars
Watch it here:
A game engine plus video model from Skywork, with an open Apache 2.0 licence.
In your videos: Named in 1 of your videos
Watch it here:
A pack of 15 to 27 coding skills, including Grill Me, which asks questions before acting.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 5 of your videos
Popularity: 302.9k stars
Watch it here:
and 1 more videos
Switch-on connections inside Hermes to tools like Canva, Notion, GitHub and NotebookLM.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Tests MCP servers. Run it with npx @modelcontextprotocol/inspector.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 10.7k stars
Watch it here:
A list with switches to turn off MCP servers you do not use.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
After each call, Hermes pulls the transcript and turns action items into tasks.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Shared memory across Hermes, Claude Code and Codex through MCP. It has a free tier.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 2 of your videos
Popularity: 67.3k stars
Watch it here:
Two notes files Hermes reads at the start of every session so it remembers you.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Metics Media had the agent build a NOTES_INDEX.md as well · Named in 9 of your videos
Watch it here:
and 5 more videos
Meta's everyday agent for online tasks like making bookings.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Meta's flagship models, Muse Spark (open) and Muse Wonder.
In your videos: Named in 3 of your videos
Watch it here:
Ask a model to write the agent's prompt or goal before you run it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Models that answer with probabilities, to help make decisions.
In your videos: Named in 1 of your videos
Watch it here:
An open-weight base model with 80B active parts. Not available yet.
In your videos: Named in 1 of your videos
Watch it here:
Two image makers that turn text into pictures.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
An open model named as Tina Huang's favourite for Hermes.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
A video generation model that needs 8xH100 chips. The name may be wrong.
Lets you train a tiny language model from scratch in PyTorch.
In your videos: Named in 1 of your videos
Watch it here:
A custom dashboard your agent builds to show tasks, content, memory, docs and agent status.
Suits: Cyborg. Cyborg runs the LAN dashboard Chris is planning.
In your videos: Named in 4 of your videos
Watch it here:
A dashboard for several Hermes agents showing tasks, health checks and live costs.
Suits: Cyborg, Forge. Cyborg runs the dashboard side and Forge runs fleet ops; matches Chris's planned agent dashboard.
In your videos: Sharbel presents it as his own tool for running a Hermes fleet · Named in 2 of your videos
Watch it here:
Mistral's models, for hosted use or small local use.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
A voice agent model that handles several speakers and can call tools.
In your videos: Named in 1 of your videos
Watch it here:
Sends one question to several models, then a lead model merges their answers.
Suits: Cerebra. Cerebra could cross-check research answers across models.
In your videos: Named in 2 of your videos
Watch it here:
A library of app screens and flows to copy design ideas from. See the Mobbin MCP.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Gives agents Mobbin's library of app screens and flows for design ideas.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Changes which AI model Hermes uses, right from the chat.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 4 of your videos
Watch it here:
The standard way agents and servers share tools, resources and prompts. It is set up in mcp.json.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Shows how much of your AI subscription usage is left.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Send each question to the smallest model that can handle it, to keep costs down.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Cleans up Mac disk space, fully uninstalls apps and analyses the disk.
In your videos: Named in 1 of your videos
Watch it here:
A tiny model that can look at webcam frames and describe what it sees.
A daily scheduled job that asks your top priority, briefs you and starts work.
Suits: Alfred, Brainiac. Alfred, the general assistant, could use it in everyday chats; Brainiac could also use it.
In your videos: Named in 4 of your videos
Watch it here:
Sends a daily summary of your calendar and email to your phone.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Several named agents with set jobs, each on its own channel and model.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 4 of your videos
Watch it here:
A task board and runtimes for agents. Setup uses API keys and hooks.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Self-hosted workflow automation for no-code agents, with Evaluations nodes and a starter kit.
Suits: Brainiac, Cerebra. Chris runs n8n for the nightly note and research workflows.
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
Nodes that score your agent's answers against test cases kept in Google Sheets.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A bundle of n8n with Ollama, Qdrant and Postgres, ready to self-host.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 15.2k stars
Watch it here:
Google's image model, used to make a daily post image.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
A self-hosted music server.
In your videos: Named in 1 of your videos
Watch it here:
A small language model.
In your videos: Named in 1 of your videos
Watch it here:
A local agent framework. The catalogue only names it.
Nvidia open models, free on OpenRouter for small side jobs your agents do.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 4 of your videos
Watch it here:
Hosts websites, and works as a connector to publish SEO content.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Starts a fresh session when the chat gets too full, which keeps costs down.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Greg Isenberg's guest says one session per topic keeps context slim · Named in 3 of your videos
Watch it here:
An AI acts as your PR manager and finds reporters.
In your videos: Named in 1 of your videos
Watch it here:
A self-hosted file cloud that works over WebDAV.
In your videos: Named in 1 of your videos
Watch it here:
Gives self-hosted apps proper domain addresses and fixes CORS errors.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A Linux system set up declaratively, using Nix flakes.
In your videos: Named in 1 of your videos
Watch it here:
Works out how to build an app just by looking at its source.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
File graph like Obsidian combined with RAG. The name may be garbled.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Handwritten notes on iPad, with transcription.
In your videos: Named in 1 of your videos
Watch it here:
Connects Hermes to NotebookLM for research and a podcast-style daily brief.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
A workspace for SOPs and task pipelines, also used as an alternative home for agent memory.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 10 of your videos
Watch it here:
and 6 more videos
Notion's built-in AI, described as a locked-in black box compared with building your own brain.
In your videos: Named in 1 of your videos
Watch it here:
A Nous Research subscription, about 20 dollars a month, giving Hermes 200+ models and hosted Hermes.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 9 of your videos
Watch it here:
and 5 more videos
An open-source skill from Sharbel for researching YouTube content, loaded automatically.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
A game engine sends a colour map and a video model draws it live.
In your videos: Named in 1 of your videos
Watch it here:
OpenAI reasoning models used to tidy up research questions before searching.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
OBS streaming with WebSocket control, fed by the Moblin iPhone camera over SRT.
In your videos: Named in 1 of your videos
Watch it here:
Note app using plain markdown files. Your agents use it as their memory, wiki and workspace.
Suits: Brainiac, Cyborg, Cerebra. Brainiac files the vault, Cyborg publishes it and Cerebra indexes it.
Watch out: Mason Fitzgerald argues file search over Obsidian is slow and costly versus RAG
In your videos: Named in 20 of your videos
Watch it here:
and 16 more videos
Database-style views over your vault notes.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A visual board in Obsidian for charts, boards and views.
Suits: Brainiac, Mantis. Brainiac files and maintains the vault and runs cron jobs, which this supports; Mantis could also use it.
In your videos: Named in 2 of your videos
Watch it here:
An Obsidian setting that lets an agent work on your vault directly.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Custom Obsidian plugin with focus, metrics, a heat map and a quick-capture inbox.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
Watch out: Sold through his community
In your videos: Named in 1 of your videos
Watch it here:
Script and skill that limit which vaults an agent is allowed to touch.
Suits: Sentinel, Brainiac. Sentinel and Brainiac could enforce the archive-folder ban with an allow-list.
In your videos: Named in 1 of your videos
Watch it here:
A built-in skill that saves and reads notes in your Obsidian vault as extra memory.
Suits: Brainiac. Brainiac files notes into the vault.
In your videos: Named in 1 of your videos
Watch it here:
Skill pack from Obsidian's CEO that teaches agents the Obsidian command line, markdown, Bases, Canvas and Defuddle.
Suits: Brainiac. Brainiac writes vault notes and needs correct wikilinks and front matter.
In your videos: Named in 2 of your videos
Popularity: 50.0k stars
Watch it here:
Paid, encrypted sync so several computers and agents share one vault.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Saves web pages and YouTube transcripts into your vault's raw folder.
Suits: Cerebra, Brainiac. Cerebra does web and source-graded research and vault indexing, which this feeds; Brainiac could also use it.
In your videos: Named in 1 of your videos
Watch it here:
An interactive world model you can move around in.
In your videos: Named in 1 of your videos
Watch it here:
Anthropic's reference examples of skills.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Splits a task between specialist agents, runs them side by side and checks the result.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Popularity: 40.2k stars
Watch it here:
Manages local models and serves them at localhost:11434. Works with Hermes and Ollama Cloud.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 9 of your videos
Popularity: 184.7k stars
Watch it here:
and 5 more videos
A Linux setup from DHH with an AI agent built in.
In your videos: Named in 3 of your videos
Popularity: 45.7k stars
Watch it here:
Turns one image plus audio into a talking video, up to 5 minutes.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Free voice model covering 600+ languages, used behind Voice Studio.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Python kit for building teams of agents, shown working with Ollama models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An OpenAI endpoint that returns simple choices, probabilities or scores instead of text.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Lets a connected app trigger the agent through MCP events.
In your videos: Named in 1 of your videos
Watch it here:
A copy of Claude Code that can run other models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Presenters say its legality is unsettled
In your videos: Named in 1 of your videos
Watch it here:
Local agent platform you install with one command. It includes a security audit command.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Alex Finn suggests it as a backup agent; Greg Isenberg's guest switched away from it · Named in 18 of your videos
Popularity: 393.0k stars
Watch it here:
and 14 more videos
A beta switch that moves strong memories into memory.md.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Asks Hermes during install to bring in an existing OpenClaw setup.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
Free, open-source terminal coding agent that works with local and hosted models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claimed 220k stars; already among Chris's built-in Hermes skills · Named in 6 of your videos
Popularity: 221.5k stars
Watch it here:
and 2 more videos
An open-source shared design canvas with an agent. Bring your own key.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Makes an interactive online classroom with slides and AI students you can practise with.
In your videos: Named in 1 of your videos
Watch it here:
Open-source video production with 12 pipelines and over 700 skill files.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Watch it here:
One API key that reaches hundreds of models, with spending limits per key, rankings and free models.
Suits: Forge. Forge manages models and each agent's OpenRouter key and budget.
In your videos: Moe Lueker uses it as a cost router to the cheapest capable model · Named in 14 of your videos
Watch it here:
and 10 more videos
Open-source alternative to Semrush and Ahrefs with real keyword and ranking data. Pay per use.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 3 of your videos
Watch it here:
Makes agent replies appear as visual screens like dashboards and forms.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Claimed 13,000 stars · Named in 1 of your videos
Watch it here:
An open chat interface for AI models that plugs into n8n and MCP connectors.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Lets you run several coding agents in one window.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Notes app built on Emacs Org files, with a web version you can host yourself.
In your videos: Named in 1 of your videos
Watch it here:
A list of other open and business models from Tina Huang's tier tour.
In your videos: Named in 1 of your videos
Watch it here:
A text screen in your terminal for viewing and managing your Docker containers.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 1.9k stars
Watch it here:
Runs Hermes, Claude Code, Codex and Pi agents like a company, with an org chart, tasks and secrets.
Suits: Forge, Cyborg. Forge runs agent ops and Cyborg dashboards; Paperclip can hire Hermes agents into one board.
In your videos: NetworkChuck says try it right now; claimed 100k+ stars · Named in 5 of your videos
Popularity: 105.5k stars
Watch it here:
and 1 more videos
Stores your documents, adds tags by rules, and lets you search them.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
An AI tool for looking up news and researching topics.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Perplexity's agent, which now runs fully on a Mac.
Watch out: Better Creating calls it expensive; about 36 GB RAM suggested
In your videos: Named in 2 of your videos
Watch it here:
Open 4B and 27B models that make decisions.
In your videos: Named in 1 of your videos
Watch it here:
A model that searches the web as it answers. Used by agents for web research.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 3 of your videos
Watch it here:
Web research tool you add to n8n agents.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
Write one skill per project and keep adding context to it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Switches the persona for one session without changing the SOUL.md file.
Suits: Mantis, Alfred. Mantis does image prompts and creative work, which this produces; Alfred could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Unreleased local app with workspaces, sessions, bots, automations and dashboards over Claude Code or Codex.
In your videos: Named in 1 of your videos
Watch it here:
Animated avatars for your bots in Hermes bot mode.
In your videos: Named in 1 of your videos
Watch it here:
A community plugin that lets Hermes run approved scripts on a real Android or mirrored iPhone.
Suits: Sentinel. Sentinel would need to lock it down before any use.
Watch out: On by default in messaging apps, so others messaging the bot could control the phone; needs Hermes 0.21.6 or newer
In your videos: Named in 1 of your videos
Watch it here:
Free iMessage link for agents, with no Mac mini needed.
In your videos: Named in 1 of your videos
Watch it here:
A very small agent harness that works well with local models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Popularity: 117.8k stars
Watch it here:
Turns product photos into videos inside ChatGPT.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
A tiny, fast text-to-speech voice that runs on your own machine.
Suits: Alfred. Alfred could voice replies cheaply, related to Chris's voice plans.
In your videos: Komputer Mechanic found it about 2x faster than Kokoro · Named in 2 of your videos
Watch it here:
Blocks edits until you approve a plan.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Simon Scrapes says the Claude Code team is considering removing it
In your videos: Named in 1 of your videos
Watch it here:
Keeps plan, findings and progress files so the agent does not lose its plan.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 27.5k stars
Watch it here:
A Microsoft tool that opens a real browser to test your app. Lighter than the MCP version.
Suits: Cyborg, Forge. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches; Forge could also use it.
In your videos: Named in 1 of your videos
Popularity: 98.3k stars
Watch it here:
Microsoft's official tool for controlling a browser using page accessibility snapshots.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Popularity: 38.0k stars
Watch it here:
Runs your own media server for films and music, with a music app.
In your videos: Named in 1 of your videos
Watch it here:
Manages the add-on packages a Python project depends on.
In your videos: Named in 1 of your videos
Watch it here:
Skill that cuts token use. One tester measured about 10 percent saved.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A dashboard for Docker containers that deploys stacks from compose files.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A database service for self-hosted apps and n8n.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
Lets you schedule social media posts ahead of time, like Buffer. The name is uncertain.
In your videos: Named in 1 of your videos
Watch it here:
Applies about 20 design principles to web pages.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Gives you ready-made project templates on a Mac so you do not start from nothing.
In your videos: Named in 1 of your videos
Watch it here:
Chat with your own PDFs on your computer, at localhost:8001.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Changing model, effort, MCPs, plugins, or compacting resets the cache and costs more.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Runs attack and repeat tests on your prompts to catch weak spots and regressions.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Popularity: 24.2k stars
Watch it here:
Software for running virtual machines, named as a home for Paperclip or Hermes.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Manages which Python versions are installed on your machine.
In your videos: Named in 1 of your videos
Watch it here:
A vector database included in the n8n self-hosted AI starter kit, for searching by meaning.
Suits: Cerebra. Cerebra's vault semantic index is planned on Qdrant.
In your videos: Named in 1 of your videos
Popularity: 34.9k stars
Watch it here:
Lines up a prompt for the agent's next turn.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
A local open model that powers Tina Huang's Hermes on a Mac Studio.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Alibaba's open models in many sizes, made for running local agents.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 11 of your videos
Watch it here:
and 7 more videos
Coding models you can run on your own machine.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A fast open image model with a Hugging Face Space to try it.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Mac app that puts several agents in one conversation with no API key.
In your videos: Named in 1 of your videos
Watch it here:
Lets a model look things up in a vector store, so it can use your private data.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 3 of your videos
Watch it here:
A vector database plus a hook that adds the 8 closest notes before Claude sees your prompt.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: Package sits behind a paid community; needs Claude Code, not Cowork or web
In your videos: Claims RAG is 5.2x faster and 3.5x cheaper than file search · Named in 1 of your videos
Watch it here:
A platform for deploying apps.
In your videos: Named in 1 of your videos
Watch it here:
Usable RAM is 75% of RAM. Divide by 0.6 to get billions of parameters that fit.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Lets you pick how hard the model should think, next to the model picker.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
A Claude desktop feature that watches you do a job and writes a small skill from it.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A cache used in Buzz's self-hosted stack.
In your videos: Named in 1 of your videos
Watch it here:
Gives Claude the look and style of known apps, so what it builds follows that design.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Saves lessons from a bot chat into its skills or memory.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Turns your corrections into a saved rule in CLAUDE.md or a skill.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An open-weight model mentioned in a news roundup.
In your videos: Named in 1 of your videos
Watch it here:
Redesign skill that reads your files to learn how your design is put together.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Open the Hermes dashboard only through a VPN or SSH tunnel, never openly on the internet.
Suits: Sentinel, Cyborg. Sentinel reviews and security-vets skills, which this supports; Cyborg could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Open-source tool that makes video from React code, with a map-animation skill.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Tristen O'Brien calls it the free-tier alternative to Claude Motion · Named in 7 of your videos
Popularity: 67.4k stars
Watch it here:
and 3 more videos
Save a session under a name and bring it back after clearing.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Sample add-on with a replay button that steps through each edit Claude just made.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Detects deepfakes and AI-made content.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Sends email from code, so an automatic newsletter can go out without you.
In your videos: Named in 1 of your videos
Watch it here:
Clears the session or summarises it so the context gets smaller.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 2 of your videos
Watch it here:
When a Telegram bot is open to everyone, allow admin commands for one Telegram ID only.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
An agent studies an app and rebuilds a feature or the whole app.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claimed 240k to 300k stars; named in seven videos · Named in 7 of your videos
Watch it here:
and 3 more videos
Dump your context, then let the agent interview you or write the best prompt itself.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 3 of your videos
Watch it here:
Undo recent turns without losing the prompt cache.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
VS Code add-on that works as an AI helper and can use MCP tools.
In your videos: Named in 1 of your videos
Watch it here:
A YAML list of routines checked every 5 minutes with a run cap, plus a read-only dashboard.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Filters terminal output before it reaches the agent's working memory.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Nate Herk's test found no savings; AI Labs' RTK skill is paid
In your videos: Named in 2 of your videos
Popularity: 84.0k stars
Watch it here:
Pick tasks that are repetitive, rules-based and give time back.
In your videos: Named in 1 of your videos
Watch it here:
Rent powerful graphics cards by the hour.
In your videos: Named in 1 of your videos
Watch it here:
A Claude Code add-on with a progress bar and a live panel of helper agents, models and cost.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A scheduled job that checks your agent's security every hour or twice a day.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 3 of your videos
Watch it here:
More than 160 skills for science research tasks.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: Video 17 reports one science skill pushes a paid product through its skill file
In your videos: Named in 2 of your videos
Watch it here:
Give agents read-only keys and database users that cannot see tables holding personal data.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
API that scrapes social media and YouTube.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Python scraper with a stealth fetcher. Offered as a free Firecrawl replacement and optional Hermes skill.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: Nuno Tavares says skip: it gets around blockers
In your videos: Claimed 87,000 stars · Named in 2 of your videos
Popularity: 87.9k stars
Watch it here:
Remote desktop from your iPad to a Mac.
In your videos: Named in 1 of your videos
Watch it here:
ByteDance's model for making video from prompts.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Watch it here:
Your agent saves repeated jobs, checklists or SOPs as its own reusable skills.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Checks a website's search ranking setup using 12 smaller skills.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Searches your old past conversations with the agent so you can find what was said.
Suits: Alfred, Brainiac. Alfred and Brainiac can recall earlier chats on demand.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Sets your home channel for scheduled results and checks whether you are the admin.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
200 free animated website backgrounds that agents can use.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
A team folder of markdown skill files that every agent reads.
Suits: Forge, Brainiac. Forge builds and runs agent operations, where this belongs; Brainiac could also use it.
In your videos: Named in 1 of your videos
Watch it here:
A short CLAUDE.md map plus index files, built from an LLM interview, so agents find context fast.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Pav Rusovs measured 30s vs 62s and 17k vs 28k tokens with the map · Named in 1 of your videos
Watch it here:
The six parts of an agent: model, tools, memory, audio, guardrails and orchestration.
In your videos: Named in 1 of your videos
Watch it here:
Builds an agent without code from your PDFs and pages, and can be put on a website.
In your videos: Named in 1 of your videos
Watch it here:
One big skill runs smaller skills in order, such as microsite, then copy, then QA.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
After each session, cleans up and improves the agent's skill collection.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
An Anthropic helper skill that builds and improves new skills.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Writes new skills and checks whether an existing skill should be extended instead.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Catalogues of shared agent skills you can browse.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
Watch out: Tina Huang avoids ClawHub skills unless from trusted authors
In your videos: Named in 2 of your videos
Watch it here:
Runs a job with and without a skill to check whether the skill really helps.
Suits: Sentinel, Forge. Sentinel and Forge could prove a skill earns its place before it ships.
In your videos: Named in 1 of your videos
Watch it here:
Scans agent skills for security problems before you install them.
Suits: Sentinel. Sentinel vets skills, so a scanner is a direct fit.
In your videos: Named in 1 of your videos
Watch it here:
Sends the model only the skills it needs, cutting skill text from about 11k to 2.3k tokens.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Try a job with no skill first, and write down only what the model gets wrong.
Suits: Forge, Sentinel. Forge and Sentinel keep skills small, matching the 15 KB guideline from video 6.
In your videos: Named in 1 of your videos
Watch it here:
Hermes catalogue of community skills you can install.
Suits: Sentinel. Sentinel should vet anything installed from the hub.
Watch out: Metics Media says to review a skill before use
In your videos: Named in 2 of your videos
Watch it here:
Team chat that agents can use, and a source for capturing notes.
In your videos: Named in 8 of your videos
Watch it here:
and 4 more videos
Community Obsidian plugin that links AI to your vault.
Suits: Brainiac, Cerebra. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cerebra could also use it.
In your videos: Named in 1 of your videos
Popularity: 5.5k stars
Watch it here:
Local memory for coding agents, based on evidence from your project.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A file that tells each agent who it is and how to behave.
Suits: Forge. Forge maintains the agents' own files.
In your videos: Jack Roberts calls it a character bible and gives a template · Named in 10 of your videos
Watch it here:
and 6 more videos
Opens sessions side by side, with one box able to drive many.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A small 2 GB image model you can run on your own computer.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Lets you nudge the agent while it works without stopping it.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Stops the current task or ends background processes.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
A payments connector, also an example of a system watched by cron.
In your videos: Named in 3 of your videos
Watch it here:
Official Stripe skill that helps Claude with payments. Install just the skill, not the whole plugin.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
One strong model plans and hands work to cheap models, each given only the tools it needs.
Suits: Forge. Forge sets models; the doc notes Chris's agents all run Haiku, so this is the part not yet used.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Small helper agents that the main agent starts for narrow jobs, then checks their work.
Suits: Forge, Cerebra. Forge and Cerebra can split work into parallel helpers.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
A desktop view to watch, stop or steer each running helper agent and change its model.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Watch it here:
Models that make music from a text prompt.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Open-source game-streaming host and client, used for remote desktop over Tailscale.
Watch out: bashbunni rejected Parsec for lacking LAN-only mode
In your videos: Named in 1 of your videos
Watch it here:
Database, login and backend memory for apps your agent builds. Has an official skill.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Nate Herk says set it read-only
In your videos: Named in 3 of your videos
Popularity: 107.9k stars
Watch it here:
Breaks a big goal into 4 to 10 actions shown on a dashboard.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Fast cloud memory that several agents can share.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 2 of your videos
Watch it here:
A set of 20+ skills for planning, building, testing, reviewing and fixing code; also has a Hermes plugin.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Nate Herk still recommends it; Teku AI cites about 287k stars · Named in 5 of your videos
Popularity: 303.9k stars
Watch it here:
and 1 more videos
Gives agents current Swift and SwiftUI documentation for building Apple apps.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A control screen and phone app for Claude Code, Codex and OpenCode.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Nimrod uses it to run 100+ projects · Named in 2 of your videos
Watch it here:
Links a few devices together, as a replacement for Tailscale.
A Tailscale beta feature that shares a folder as a network drive.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Quickly sends files between your Tailscale devices.
In your videos: Named in 1 of your videos
Watch it here:
A private, encrypted network so your agents can reach and log into all your devices.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 8 of your videos
Popularity: 36.3k stars
Watch it here:
and 4 more videos
Shares a local port to your own Tailscale network with a web address.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Breaks a product requirements document into tasks, using 36 MCP tools.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
The most recommended phone chat for Hermes, with BotFather bots, QR setup and group topics as rooms.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Alex Finn recommends it for deep mobile work · Named in 24 of your videos
Watch it here:
and 20 more videos
Turns photos taken from several angles into a detailed 3D model.
In your videos: Named in 1 of your videos
Watch it here:
Adds a real web browser beside the chat, including GitHub pull requests.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
OpenAI model that turns notes into searchable numbers. Suggested for a vault search, about $5 credit.
Suits: Cerebra. Cerebra's semantic vault index needs an embedding model.
In your videos: Named in 1 of your videos
Watch it here:
A teleprompter on your Mac that shows your script while you read it.
In your videos: Named in 1 of your videos
Watch it here:
Turns a prompt into STL files for 3D printing.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Nuno Tavares says skip it: blocked on Windows 11
In your videos: Named in 2 of your videos
Watch it here:
Google's model for forecasting numbers over time.
In your videos: Named in 1 of your videos
Watch it here:
Tiny story models small enough to run on an ESP32-S3 chip.
Daily commands: one pulls calendar and inbox to set three priorities, another logs your day by voice.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Managed services that run open AI models for you through an API.
In your videos: Named in 1 of your videos
Watch it here:
Have the agent measure and cut its own usage on each task.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A paste-in prompt that checks Claude Code context, deferral, memory files, cache and schedules.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A default mode that loads only a tool index, roughly 85% cheaper.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Turns off tools and built-in skills you do not use, to cut cost and wrong tool loops.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Keep decisions and working docs, then turn them into lessons in the brain.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A front door that gives the Hermes dashboard HTTPS in Hostinger's Docker template.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Drafts and schedules X posts using an API key. A typefully-drafts skill exists.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Basic server lockdown: a sudo user, ed25519 SSH keys, and a firewall allowing only SSH, 80 and 443.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Website design skill you install from GitHub into Codex or Claude.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
A cheap open-weight model at $0.15 in and $0.50 out.
In your videos: Named in 1 of your videos
Watch it here:
Self-hosted analytics for seeing who visits your website.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
Watch out: Default login admin/umami must be changed
In your videos: Named in 1 of your videos
Watch it here:
API for LinkedIn and Instagram DMs, WhatsApp, Messenger, Telegram and email, about $5 per account.
In your videos: Named in 1 of your videos
Watch it here:
A game engine that an agent can drive or import generated worlds into.
In your videos: Named in 2 of your videos
Watch it here:
Fine-tunes open models, either on your own computer or in Colab.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Posts to many social platforms at once, with a command line and a skill.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
A model option you can use in Hermes.
In your videos: Named in 1 of your videos
Watch it here:
Shows which skill, tool or agent is using up your plan.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
A fast tool for starting, adding to and running Python projects.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 90.5k stars
Watch it here:
Pav Rusovs' skills: nightly map rebuild, broken-link check, Mac cleanup and a media skill.
Suits: Brainiac. Brainiac maintains the vault and could lint broken links nightly.
In your videos: Named in 1 of your videos
Watch it here:
Three models that make video from prompts.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Free hosting for dashboards and sites, with a Doppler sync.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 4 of your videos
Watch it here:
A skill that finds the best skill for your task, listed on skills.sh.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Popularity: 36.3k stars
Watch it here:
Claude plugin used in a demo to publish a design to Vercel.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
A text-to-speech model. The caption was garbled.
In your videos: Named in 1 of your videos
Watch it here:
YouTube demand research, used by a YouTube agent.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
A spreadsheet-style explorer for CSV and other structured data, inside the terminal.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Talk to Hermes out loud and use a wake word to hand it tasks.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Alex Finn says ChatGPT voice mode needs an OpenAI API key, about 5 cents a minute
In your videos: Named in 3 of your videos
Watch it here:
Send voice notes from your phone that an agent on a home machine turns into text and acts on.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Free local voice cloning, scripts, dubbing and transcription. An alternative to ElevenLabs.
Suits: Mantis, Alfred. Mantis does image prompts and creative work, which this produces; Alfred could also use it.
In your videos: Claimed 61,000 stars · Named in 4 of your videos
Watch it here:
A doc built by interview about how you think, plus a voice guide per medium with banned words.
Suits: Alfred, Mantis. Alfred, the general assistant, could use it in everyday chats; Mantis could also use it.
In your videos: Named in 1 of your videos
Watch it here:
A small toolbox of handy tools that lives in the Mac menu bar.
In your videos: Named in 1 of your videos
Watch it here:
A mid-tier open video model. Slow on a Mac Studio.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Starter kit with a wiki_tool.py script, ingest, query, maintain and lint skills, and note templates.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
Watch out: Starter kit is for members only
In your videos: Named in 1 of your videos
Watch it here:
A terminal built for coding with AI help.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Builds calculators and dashboards from a plain description.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
A file-sharing method chosen over SSHFS, NFS and SMB, run with a Docker WebDAV server.
In your videos: Named in 1 of your videos
Watch it here:
An agent watches sign-ups and pages through webhooks and controls payments.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A messaging channel for Hermes and OpenClaw.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
An open model that turns speech into text, so you can talk to your agents.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Instructions tell agents to read the wiki first and drop notes into an inbox that a script files.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
An AI code editor that can use MCP tools.
In your videos: Named in 1 of your videos
Watch it here:
A voice dictation app.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Better Creating says it stopped working offline
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
Manages Git worktrees, which suits running several Claude Code agents at the same time.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Tencent's world model that remembers where the camera has been.
In your videos: Named in 1 of your videos
Watch it here:
Free website that makes random manga-style maps and cities in your browser.
In your videos: Named in 1 of your videos
Watch it here:
Spots objects in camera images on a phone using the Ultralytics app.
Anthropic add-on where a second agent points out things you may have missed.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
Watch out: Uses extra tokens
In your videos: Named in 1 of your videos
Watch it here:
Replaces the default YouTube skill with transcripts, channels and search, with no API key.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Turns YouTube videos into markdown notes using the Karpathy wiki method.
Suits: Cerebra, Brainiac. Cerebra does web and source-graded research and vault indexing, which this feeds; Brainiac could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Reads, edits and converts YAML, JSON, XML, CSV and TOML files.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 15.9k stars
Watch it here:
Standalone MCP server that works without Zapier, using your own tokens.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Do actions across Zapier's 9,000 apps by chatting, with one central log.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 4 of your videos
Watch it here:
Reusable scripts with saved sign-ins. Cheaper than MCP for repeated tasks.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Corbin claims about 210x cost cut versus MCP in a separate video · Named in 3 of your videos
Watch it here:
Memory layer that tracks how facts change over time.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
A library of interface parts that Claude can pull from.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
The 10 named in the most videos, of everything. Ranked by how many of your 105 videos named each one. Tinted by chapter colour.
The main program that runs your always-on agents. It has memory, skills, scheduled jobs and over 20 chat channels.
Named in 56 of your videos
Suits: Alfred, Brainiac, Cerebra. It is the runtime all seven of Chris's agents already use.
In your videos: Alex Finn calls it the greatest AI tool ever made · Named in 56 of your videos
Popularity: 253.6k stars
Watch it here:
and 52 more videos
Anthropic's coding agent for the terminal, with skills, plugins, hooks, MCP and helper agents.
Named in 39 of your videos
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Pav Rusovs chose it over Hermes, citing Anthropic rules on subscriptions in third-party harnesses · Named in 39 of your videos
Popularity: 153.3k stars
Watch it here:
and 35 more videos
OpenAI's coding agent, in the terminal or the ChatGPT desktop app.
Named in 35 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Samin Yasar uses Codex as the doctor that fixes Hermes · Named in 35 of your videos
Popularity: 128.1k stars
Watch it here:
and 31 more videos
Anthropic's top model line. Opus 5.5 is the one Anthropic suggests as your default.
Named in 24 of your videos
Suits: Forge. Forge could use it as a strong orchestrator; the playlist-1 doc notes Chris's agents all run Haiku today.
Watch out: Alex Finn says it costs him about 40 USD a day; Moe Lueker says avoid it for routine tasks
In your videos: Named in 24 of your videos
Watch it here:
and 20 more videos
The most recommended phone chat for Hermes, with BotFather bots, QR setup and group topics as rooms.
Named in 24 of your videos
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Alex Finn recommends it for deep mobile work · Named in 24 of your videos
Watch it here:
and 20 more videos
Note app using plain markdown files. Your agents use it as their memory, wiki and workspace.
Named in 20 of your videos
Suits: Brainiac, Cyborg, Cerebra. Brainiac files the vault, Cyborg publishes it and Cerebra indexes it.
Watch out: Mason Fitzgerald argues file search over Obsidian is slow and costly versus RAG
In your videos: Named in 20 of your videos
Watch it here:
and 16 more videos
Local agent platform you install with one command. It includes a security audit command.
Named in 18 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Alex Finn suggests it as a backup agent; Greg Isenberg's guest switched away from it · Named in 18 of your videos
Popularity: 393.0k stars
Watch it here:
and 14 more videos
Cheap open models that use tools well. Popular choice to power Hermes.
Named in 16 of your videos
Suits: Alfred, Cerebra. Alfred, the general assistant, could use it in everyday chats; Cerebra could also use it.
In your videos: Metics Media picks V4 Flash as a cheap main model · Named in 16 of your videos
Watch it here:
and 12 more videos
One API key that reaches hundreds of models, with spending limits per key, rankings and free models.
Named in 14 of your videos
Suits: Forge. Forge manages models and each agent's OpenRouter key and budget.
In your videos: Moe Lueker uses it as a cost router to the cheapest capable model · Named in 14 of your videos
Watch it here:
and 10 more videos
Runs a task on a timer by itself and sends the result to a chat channel.
Named in 13 of your videos
Suits: Brainiac. Brainiac runs Chris's cron work.
In your videos: Alex Finn says to prune and pause old cron jobs to improve speed and cost · Named in 13 of your videos
Watch it here:
and 9 more videos
Programs and services. Ranked by how many of your 105 videos named each one. Tinted by chapter colour.
The main program that runs your always-on agents. It has memory, skills, scheduled jobs and over 20 chat channels.
Named in 56 of your videos
Suits: Alfred, Brainiac, Cerebra. It is the runtime all seven of Chris's agents already use.
In your videos: Alex Finn calls it the greatest AI tool ever made · Named in 56 of your videos
Popularity: 253.6k stars
Watch it here:
and 52 more videos
The most recommended phone chat for Hermes, with BotFather bots, QR setup and group topics as rooms.
Named in 24 of your videos
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Alex Finn recommends it for deep mobile work · Named in 24 of your videos
Watch it here:
and 20 more videos
Note app using plain markdown files. Your agents use it as their memory, wiki and workspace.
Named in 20 of your videos
Suits: Brainiac, Cyborg, Cerebra. Brainiac files the vault, Cyborg publishes it and Cerebra indexes it.
Watch out: Mason Fitzgerald argues file search over Obsidian is slow and costly versus RAG
In your videos: Named in 20 of your videos
Watch it here:
and 16 more videos
Local agent platform you install with one command. It includes a security audit command.
Named in 18 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Alex Finn suggests it as a backup agent; Greg Isenberg's guest switched away from it · Named in 18 of your videos
Popularity: 393.0k stars
Watch it here:
and 14 more videos
One API key that reaches hundreds of models, with spending limits per key, rankings and free models.
Named in 14 of your videos
Suits: Forge. Forge manages models and each agent's OpenRouter key and budget.
In your videos: Moe Lueker uses it as a cost router to the cheapest capable model · Named in 14 of your videos
Watch it here:
and 10 more videos
A chat app where your Hermes agents live, with one channel per project.
Named in 12 of your videos
Suits: Alfred. Chris's agents already talk in Discord channels.
In your videos: Named in 12 of your videos
Watch it here:
and 8 more videos
A rented server with one-click Hermes, Dokploy and n8n templates and the Kodee assistant.
Named in 12 of your videos
Suits: Forge. Forge runs hosting, but Chris already hosts on Coolify, so this is only a comparison.
Watch out: Lead Gen Jay calls Hostinger servers weak
In your videos: Named in 12 of your videos
Watch it here:
and 8 more videos
xAI's agent app. Each bot gets its own cloud computer, routines and reports.
Named in 10 of your videos
In your videos: Named in 10 of your videos
Watch it here:
and 6 more videos
A workspace for SOPs and task pipelines, also used as an alternative home for agent memory.
Named in 10 of your videos
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 10 of your videos
Watch it here:
and 6 more videos
A desktop app for using Hermes: chats, models, profiles, scheduled jobs, skills and pinned sessions.
Named in 9 of your videos
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 9 of your videos
Watch it here:
and 5 more videos
Built-in switches inside Hermes and Claude. Ranked by how many of your 105 videos named each one. Tinted by chapter colour.
Runs a task on a timer by itself and sends the result to a chat channel.
Named in 13 of your videos
Suits: Brainiac. Brainiac runs Chris's cron work.
In your videos: Alex Finn says to prune and pause old cron jobs to improve speed and cost · Named in 13 of your videos
Watch it here:
and 9 more videos
A task board where work moves from new to done and gets shared out between helper agents.
Named in 10 of your videos
Suits: Forge, Brainiac. Forge runs builds and ops; Brainiac could own filing tasks on the board.
In your videos: Named in 10 of your videos
Watch it here:
and 6 more videos
A file that tells each agent who it is and how to behave.
Named in 10 of your videos
Suits: Forge. Forge maintains the agents' own files.
In your videos: Jack Roberts calls it a character bible and gives a template · Named in 10 of your videos
Watch it here:
and 6 more videos
Two notes files Hermes reads at the start of every session so it remembers you.
Named in 9 of your videos
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Metics Media had the agent build a NOTES_INDEX.md as well · Named in 9 of your videos
Watch it here:
and 5 more videos
Separate Hermes setups, each with its own keys, memory, scheduled jobs, skills and personality.
Named in 6 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Alex Finn recommends at least two profiles so one can fix the other · Named in 6 of your videos
Watch it here:
and 2 more videos
A web page to see and manage your Hermes sessions, models, jobs, skills, keys, channels and logs.
Named in 6 of your videos
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Keeps the agent working until your stated goal is done, with a limit on turns.
Named in 5 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Jack Roberts says a goal has a 20-turn budget by default
In your videos: Sharbel says to have Hermes write the goal prompt first (meta-prompting) · Named in 5 of your videos
Watch it here:
and 1 more videos
Searches your old past conversations with the agent so you can find what was said.
Named in 5 of your videos
Suits: Alfred, Brainiac. Alfred and Brainiac can recall earlier chats on demand.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Small helper agents that the main agent starts for narrow jobs, then checks their work.
Named in 5 of your videos
Suits: Forge, Cerebra. Forge and Cerebra can split work into parallel helpers.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Changes which AI model Hermes uses, right from the chat.
Named in 4 of your videos
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 4 of your videos
Watch it here:
Ways of working worth copying. Ranked by how many of your 105 videos named each one. Tinted by chapter colour.
A file of instructions Claude reads. Keep it a short map to other files, and check it now and then.
Named in 7 of your videos
Suits: Forge, Brainiac. Forge builds and runs agent operations, where this belongs; Brainiac could also use it.
Watch out: Mason Fitzgerald calls it a soft instruction that is not enforced
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
One shared instruction file for agents. Claude Code now reads it, and it can sit in any folder.
Named in 6 of your videos
Suits: Forge, Brainiac. Forge builds and runs agent operations, where this belongs; Brainiac could also use it.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
A pattern where an LLM turns raw sources into a linked markdown wiki, with ingest, query and lint steps.
Named in 6 of your videos
Suits: Brainiac, Cerebra. Brainiac maintains the vault and Cerebra indexes it; this is the method behind both.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Your agent saves repeated jobs, checklists or SOPs as its own reusable skills.
Named in 6 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
One strong model plans and hands work to cheap models, each given only the tools it needs.
Named in 5 of your videos
Suits: Forge. Forge sets models; the doc notes Chris's agents all run Haiku, so this is the part not yet used.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
A searchable wiki of decisions, logs and research that your agent keeps up to date so it stays consistent.
Named in 4 of your videos
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 4 of your videos
Watch it here:
A custom dashboard your agent builds to show tasks, content, memory, docs and agent status.
Named in 4 of your videos
Suits: Cyborg. Cyborg runs the LAN dashboard Chris is planning.
In your videos: Named in 4 of your videos
Watch it here:
Several named agents with set jobs, each on its own channel and model.
Named in 4 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 4 of your videos
Watch it here:
A daily scheduled job that asks your top priority, briefs you and starts work.
Named in 4 of your videos
Suits: Alfred, Brainiac. Alfred, the general assistant, could use it in everyday chats; Brainiac could also use it.
In your videos: Named in 4 of your videos
Watch it here:
Each agent writes down what it did, or adds it to today's daily note, before replying.
Named in 3 of your videos
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 3 of your videos
Watch it here:
The AI brains. Ranked by how many of your 105 videos named each one. Tinted by chapter colour.
Anthropic's top model line. Opus 5.5 is the one Anthropic suggests as your default.
Named in 24 of your videos
Suits: Forge. Forge could use it as a strong orchestrator; the playlist-1 doc notes Chris's agents all run Haiku today.
Watch out: Alex Finn says it costs him about 40 USD a day; Moe Lueker says avoid it for routine tasks
In your videos: Named in 24 of your videos
Watch it here:
and 20 more videos
Cheap open models that use tools well. Popular choice to power Hermes.
Named in 16 of your videos
Suits: Alfred, Cerebra. Alfred, the general assistant, could use it in everyday chats; Cerebra could also use it.
In your videos: Metics Media picks V4 Flash as a cheap main model · Named in 16 of your videos
Watch it here:
and 12 more videos
Anthropic's middle model, called the consensus best choice to power Hermes.
Named in 13 of your videos
Suits: Forge, Cerebra. Forge builds and runs agent operations, where this belongs; Cerebra could also use it.
In your videos: Dan Martell: Sonnet to run agents, Opus to build them · Named in 13 of your videos
Watch it here:
and 9 more videos
OpenAI's top model, strong at using computers and coding. It powers ChatGPT Dots.
Named in 13 of your videos
In your videos: Alex Finn says only Opus 5.5 is ahead of it · Named in 13 of your videos
Watch it here:
and 9 more videos
Anthropic's strongest tier, named as an advisor model above Opus.
Named in 12 of your videos
Watch out: Jason Lee hit usage limits with it
In your videos: Named in 12 of your videos
Watch it here:
and 8 more videos
ChatGPT-plan models you can use in Hermes by logging in, with no API cost.
Named in 11 of your videos
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Alex Finn says 5.5 is the first usable cheap option via a ChatGPT subscription · Named in 11 of your videos
Watch it here:
and 7 more videos
Moonshot's open model, used as a cheap option to power Hermes.
Named in 11 of your videos
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 11 of your videos
Watch it here:
and 7 more videos
Alibaba's open models in many sizes, made for running local agents.
Named in 11 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 11 of your videos
Watch it here:
and 7 more videos
An open model from Z.AI. It is the cheapest option Alex Finn names.
Named in 9 of your videos
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 9 of your videos
Watch it here:
and 5 more videos
Anthropic's fast, cheap model. Good for small coding helper agents where cost matters.
Named in 8 of your videos
Suits: Alfred, Brainiac. All seven agents run on Haiku today.
In your videos: Sharbel suggests Haiku sub-agents to protect the main cache · Named in 8 of your videos
Watch it here:
and 4 more videos
Connect an agent to other tools. Ranked by how many of your 105 videos named each one. Tinted by chapter colour.
Makes images and video with major models through an MCP connector and command line, including face-consistent Soul.
Named in 6 of your videos
Suits: Mantis. Mantis makes images and could generate them directly.
In your videos: Tristen O'Brien adds it to Claude as a custom connector · Named in 6 of your videos
Watch it here:
and 2 more videos
Meeting notes app with no bot in the call. Its MCP lets agents chat with your meeting history.
Named in 6 of your videos
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Connects agents to over 1,000 apps like Gmail, Slack and Notion through sign-in, and has a command line.
Named in 5 of your videos
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 5 of your videos
Popularity: 30.7k stars
Watch it here:
and 1 more videos
Turns websites into clean markdown. It can scrape, crawl and get past bot detection.
Named in 4 of your videos
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: The Next New Thing offers Scrapling as a free replacement
In your videos: Named in 4 of your videos
Popularity: 194.9k stars
Watch it here:
Turns your code into a lasting knowledge graph so the agent uses fewer tokens exploring.
Named in 4 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claims 99% (Sharbel) or 120x (Teku AI) fewer tokens; about 45k stars · Named in 4 of your videos
Popularity: 47.2k stars
Watch it here:
Memory service shared across agents, with dreaming. Check it with the command hermes honcho status.
Named in 4 of your videos
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 4 of your videos
Popularity: 7.8k stars
Watch it here:
Do actions across Zapier's 9,000 apps by chatting, with one central log.
Named in 4 of your videos
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 4 of your videos
Watch it here:
Lets an agent control Blender to make 3D models and games.
Named in 3 of your videos
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Popularity: 32.3k stars
Watch it here:
Chrome extension that lets ChatGPT or Codex control your browser and computer.
Named in 3 of your videos
In your videos: Named in 3 of your videos
Watch it here:
Anthropic's multi-agent review that checks your code for security problems.
Named in 2 of your videos
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 2 of your videos
Watch it here:
Code projects on GitHub. Ranked by how many of your 105 videos named each one. Tinted by chapter colour.
An agent studies an app and rebuilds a feature or the whole app.
Named in 7 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claimed 240k to 300k stars; named in seven videos · Named in 7 of your videos
Watch it here:
and 3 more videos
A pack of 15 to 27 coding skills, including Grill Me, which asks questions before acting.
Named in 5 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 5 of your videos
Popularity: 302.9k stars
Watch it here:
and 1 more videos
Long-term agent memory that can retain, recall and reflect. A Hermes memory option that runs in Docker.
Named in 4 of your videos
Suits: Brainiac, Cerebra. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cerebra could also use it.
Watch out: Playlist-1 doc notes it overlaps Chris's Qdrant plan
In your videos: Named in 4 of your videos
Popularity: 44.6k stars
Watch it here:
Copies the format of a viral video.
Named in 4 of your videos
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Nuno Tavares says skip: needs paid services
In your videos: Named in 4 of your videos
Watch it here:
About 24 senior-engineer skills with commands to spec, plan, build, test, review and ship.
Named in 3 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Popularity: 110.0k stars
Watch it here:
Garry Tan's chain of role skills for planning, reviewing and shipping software.
Named in 3 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: AI Edge cites 133k stars · Named in 3 of your videos
Popularity: 137.3k stars
Watch it here:
A Claude Code script that removes ads from a TCL smart TV.
Named in 3 of your videos
In your videos: Claimed 22,000 stars · Named in 3 of your videos
Watch it here:
A pack of about 20 to 49 skills for SEO, ads, copywriting and conversion.
Named in 3 of your videos
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Popularity: 56.3k stars
Watch it here:
Open-source video production with 12 pipelines and over 700 skill files.
Named in 3 of your videos
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Watch it here:
Open-source alternative to Semrush and Ahrefs with real keyword and ranking data. Pay per use.
Named in 3 of your videos
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 3 of your videos
Watch it here:
Add-ons that teach an agent a new job. Ranked by how many of your 105 videos named each one. Tinted by chapter colour.
The agent writes an HTML page that turns into an MP4 video.
Named in 7 of your videos
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Needs Node 22
In your videos: Named in 7 of your videos
Popularity: 71.8k stars
Watch it here:
and 3 more videos
Open-source tool that makes video from React code, with a map-animation skill.
Named in 7 of your videos
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Tristen O'Brien calls it the free-tier alternative to Claude Motion · Named in 7 of your videos
Popularity: 67.4k stars
Watch it here:
and 3 more videos
A set of 20+ skills for planning, building, testing, reviewing and fixing code; also has a Hermes plugin.
Named in 5 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Nate Herk still recommends it; Teku AI cites about 287k stars · Named in 5 of your videos
Popularity: 303.9k stars
Watch it here:
and 1 more videos
Lets agents read sites like X, Reddit, YouTube and GitHub without paying for API fees.
Named in 4 of your videos
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Teku AI cites about 81k stars and says it gets past 403 bot blocks · Named in 4 of your videos
Popularity: 101.3k stars
Watch it here:
Strips a web page down to clean readable text, which saves on tokens.
Named in 4 of your videos
Suits: Cerebra, Brainiac. Cerebra fetches pages for research and Brainiac clips into the vault; both save tokens.
In your videos: Named in 4 of your videos
Popularity: 9.4k stars
Watch it here:
Rewrites text to remove the usual signs that an AI wrote it.
Named in 4 of your videos
Suits: Alfred, Mantis. Alfred and Mantis write text for Chris; it is already built into Hermes.
Watch out: Teku AI says it is token heavy; Sharbel notes the skill file is 29.6 KB
In your videos: Playlist-1 doc notes it is already built into Hermes · Named in 4 of your videos
Popularity: 60.1k stars
Watch it here:
Researches a topic from the last 30 days across Reddit, X, YouTube, GitHub and the web.
Named in 3 of your videos
Suits: Cerebra. Cerebra does web research and this gives her fresh multi-platform sentiment.
In your videos: AI Edge notes 140 contributors; Futurepedia lists it first · Named in 3 of your videos
Popularity: 65.2k stars
Watch it here:
An ultra-short answer mode that claims about 65% fewer output tokens.
Named in 3 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Nate Herk's test found only about 8.5% savings
In your videos: Teku AI cites about 100k stars · Named in 3 of your videos
Popularity: 114.0k stars
Watch it here:
A multi-step research process that uses several models at once.
Named in 2 of your videos
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
Hermes skill where a librarian bot files reports into your Obsidian wiki and asks you to approve reviews.
Named in 2 of your videos
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 2 of your videos
Watch it here:
Things you type commands for. Ranked by how many of your 105 videos named each one. Tinted by chapter colour.
Anthropic's coding agent for the terminal, with skills, plugins, hooks, MCP and helper agents.
Named in 39 of your videos
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Pav Rusovs chose it over Hermes, citing Anthropic rules on subscriptions in third-party harnesses · Named in 39 of your videos
Popularity: 153.3k stars
Watch it here:
and 35 more videos
OpenAI's coding agent, in the terminal or the ChatGPT desktop app.
Named in 35 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Samin Yasar uses Codex as the doctor that fixes Hermes · Named in 35 of your videos
Popularity: 128.1k stars
Watch it here:
and 31 more videos
Manages local models and serves them at localhost:11434. Works with Hermes and Ollama Cloud.
Named in 9 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 9 of your videos
Popularity: 184.7k stars
Watch it here:
and 5 more videos
Runs several apps and agents separately on one machine. Docker Compose handles whole stacks.
Named in 6 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Free, open-source terminal coding agent that works with local and hosted models.
Named in 6 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claimed 220k stars; already among Chris's built-in Hermes skills · Named in 6 of your videos
Popularity: 221.5k stars
Watch it here:
and 2 more videos
A very small agent harness that works well with local models.
Named in 3 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 3 of your videos
Popularity: 117.8k stars
Watch it here:
Reusable scripts with saved sign-ins. Cheaper than MCP for repeated tasks.
Named in 3 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Corbin claims about 210x cost cut versus MCP in a separate video · Named in 3 of your videos
Watch it here:
Lets an agent use Drive, Gmail, Calendar, Docs, Sheets and Slides, with over 40 skills.
Named in 2 of your videos
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Popularity: 30.4k stars
Watch it here:
Filters terminal output before it reaches the agent's working memory.
Named in 2 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Nate Herk's test found no savings; AI Labs' RTK skill is paid
In your videos: Named in 2 of your videos
Popularity: 84.0k stars
Watch it here:
Shows where your Claude tokens go and flags MCP servers you never use.
Named in 1 of your videos
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
35 warnings from your videos, worst first inside each group. Each one says why, and what to do instead.
Tokens, models and subscriptions: the bill comes from where work is sent and how much context rides along.
Why: Moe Lueker routes each task to the cheapest capable model and keeps Opus and Fable for hard work. Alex Finn puts his own Claude Opus use at about 40 US dollars a day.
Do this instead: Use a cheap main model for routine work and escalate to a stronger one only when needed; Simon Scrapes's /advisor shows the pattern.
Said by Moe Lueker, Alex Finn, Simon Scrapes
Why: Jack Roberts and Metics Media both point to OpenRouter's spend limits, which can be set per key, so one runaway agent cannot drain the shared budget.
Do this instead: Give each agent its own OpenRouter key with its own limit.
Said by Jack Roberts, Metics Media
Why: Python Simplified shows a Hermes cron job with a no-agent checkbox that runs the script and never calls the model, and an auxiliary model setting that sends background work to a free OpenRouter model.
Do this instead: Tick no-agent for script-only cron jobs and set a cheap auxiliary model.
Said by Python Simplified
Why: Alex Finn says pruning and pausing old cron jobs improves speed and cost.
Do this instead: List each job's purpose, pause what nobody reads, and archive rather than delete.
Said by Alex Finn
Why: Caveman claims about 65 per cent fewer output tokens, but Nate Herk's own test found about 8.5 per cent, and he measured no saving from RTK. The codebase-memory claim of 99 per cent comes from Sharbel A. repeating the project.
Do this instead: Measure on one agent for a week before it becomes a fleet rule.
Said by Nate Herk, Sharbel A., Teku AI
Why: Sharbel A. explains that history is re-sent every turn, and that changing model, effort, MCP or plugin toggles, or compacting, resets the prompt cache and costs more. Jay E of RoboNuggets shows Cache Tax, which warns before a send once the cache has gone cold.
Do this instead: One session per topic, start fresh with /new when context fills, and avoid mid-session toggles.
Said by Sharbel A., Greg Isenberg and Alex Finn, Jay E
Why: Greg Isenberg's guest says switching off some of Hermes's roughly 150 built-in skills saves cost. Sharbel A. says to turn off unused MCP servers, and Wanderloots prunes each bot's tool toggles to stop tool loops.
Do this instead: Audit each agent's skills, toolsets and MCP list and keep only what its lane uses.
Said by Alex Finn, Sharbel A., Wanderloots
Workflow traps: too many agents, borrowed benchmarks, paywalled freebies and promises of easy money.
Why: Dan Martell's AGENT framework gives each agent one lane, a one-sentence definition of done and more freedom only as it earns it, and his Rule of R asks whether a task is repetitive, rules-based and worth the time.
Do this instead: Write each agent's lane and done-test before adding another.
Said by Dan Martell
Why: Wanderloots caps turns in a shared room so a human must step in. Jack Roberts says a Hermes goal has a 20-turn budget by default.
Do this instead: Set a turn cap on every group room and goal, then decide as a human.
Said by Wanderloots, Jack Roberts
Why: Nuno Tavares says to skip Scrapling even though it shows about 87,900 GitHub stars. Agent Reach shows about 101,300 and is praised by Teku AI. Stars were checked on 25 Oct 2026 and measure attention only.
Do this instead: Review every skill whatever its stars, and log Sentinel's verdict in the vault.
Said by Nuno Tavares, Teku AI
Why: NetworkChuck loves Paperclip, which brings an org chart, secrets and routines of its own. Metics Media, Alex Finn and Tina Huang run multi-agent work on Hermes's built-in Kanban board.
Do this instead: Try the native Kanban board first and add a meta layer only if it falls short.
Said by NetworkChuck, Metics Media, Alex Finn
Why: Komputer Mechanic's free-prompts title leads to a $10-a-month membership. Mason Fitzgerald's RAG package, Julian Goldie's Agent OS and AI Labs's RTK skill sit behind paid communities.
Do this instead: Check what is actually free before investing time, and copy the idea instead of buying the product.
Said by Komputer Mechanic, Mason Fitzgerald, Julian Goldie, AI Labs
Why: Dubibubi let an agent run a digital-product Etsy shop for 10 days and it made 4.55 dollars. The Next New Thing's pay-for-itself episode is a Zapier-sponsored roundup. The one lasting lesson was cutting tokens per listing from 4 per cent to 0.33 per cent of a weekly limit.
Do this instead: Judge each agent by hours saved on a named task, not by promised income.
Said by Dubibubi, The Next New Thing
Why: The reports are built from auto-generated captions and flag many model names as garbled; GPT-6 Astra came through as Chief 6 Astra, and Sol 5.6 is likely a mishearing. Several videos are sponsor-shaped, such as the Zapier and OpenAI segments.
Do this instead: Confirm any model name against OpenRouter's model list before moving an agent.
Said by Julian Goldie, Lead Gen Jay
Skills, plugins and MCP servers are text and code someone else wrote; treat each one as untrusted until proven otherwise.
Why: Futurepedia shows skills going in by pasting a link, but Tina Huang avoids ClawHub skills unless the author is trusted, and Metics Media says to review anything from the Hermes Skills Hub before use.
Do this instead: Have Sentinel read the skill file line by line first, and prefer authors you can name.
Said by Tina Huang, Metics Media, Futurepedia
Why: A skill is instructions the agent obeys, so it can carry a hidden push. The Next New Thing reported a science skill that steers users to a paid product from inside its own skill file.
Do this instead: Run a prompt-injection check before install: Jack Roberts demos Bouncer for Hermes, and Futurepedia shows Nvidia's Skill Inspector.
Said by The Next New Thing, Jack Roberts, Futurepedia
Why: Julian Goldie says the community plugin is on by default in messaging apps, so anyone who can message your bot could drive the phone. The plugin's author is given only with low confidence in the report and no repo was named.
Do this instead: Turn the phone tool off per platform with the hermes tools command, or skip the plugin.
Said by Julian Goldie
Why: Nate Herk tested over a hundred repos for Claude and says to skip Claude Mem because it sends data by default, and Context Mode because of a usage-limit bug.
Do this instead: Test any memory or context plugin on a throwaway setup and read what it sends before it touches real data.
Said by Nate Herk
Why: Nuno Tavares says to skip Scrapling, a free Firecrawl replacement, because it gets around blockers. Teku AI praises Agent Reach for getting past 403 bot blocks, which is the same trait.
Do this instead: Use an approved scraper for research jobs and let Sentinel decide whether a bypass tool is acceptable at all.
Said by Nuno Tavares, Teku AI
Why: The pick-lists name packs of 49 marketing skills, 292 skills in ECC, 279 role agents and 700+ skill files in Open Montage. AI Labs shows a retriever cutting skill context from about 11,000 to 2,300 tokens, which shows what an unfiltered pile costs.
Do this instead: Install the one or two skills a given agent needs and switch off the rest.
Said by Sharbel A., The Next New Thing, AI Edge, AI Labs
Why: Teku AI calls Humanizer token heavy and Sharbel A. notes its skill file is 29.6 KB; a playlist-one note in the catalogue says it is already built into Hermes.
Do this instead: Check the built-in list first, and use Simon Scrapes's test: run the job with and without the skill and keep it only if it helps.
Said by Teku AI, Sharbel A., Simon Scrapes
Security mistakes on a VPS full of agents: who can talk to them, what they can touch, and what faces the internet.
Why: Python Simplified shows that a single environment variable, TELEGRAM_ALLOW_ALL_USERS set to true, opens the bot to everyone, so anyone can chat with the agent.
Do this instead: Restrict admin commands to your own ID in the Hermes config and check who the bot thinks you are.
Said by Python Simplified
Why: Jack Roberts says keys should never go into a conversation, where they linger in history. Chris's own no-secrets-in-Discord rule is the same idea.
Do this instead: Set keys as environment variables; Lead Gen Jay keeps tokens in Doppler, a central store.
Said by Jack Roberts, Lead Gen Jay
Why: Nimrod gives agents scoped read-only keys in Jira, GitLab and GitHub, read-only database users that cannot see personal-data tables, and a small local MCP proxy exposing only allowed routes. Nate Herk says to make Supabase read-only.
Do this instead: One scoped key per agent, read-only by default, write access added only when a lane needs it.
Said by Nimrod, Nate Herk
Why: Sharbel A. sets Hermes to ask before spending, sending or deleting, and Jack Roberts says not to let it send email. Pav Rusovs keeps a locked-down settings file that blocks send, share and delete.
Do this instead: Drafts only for email, approval gates for anything destructive, and a cap on how often a scheduled routine may run.
Said by Sharbel A., Jack Roberts, Pav Rusovs
Why: Julian Goldie lists a dashboard behind a VPN or SSH tunnel as the safe way to reach an agent remotely. Alex Finn, Nimrod and Pav Rusovs use Tailscale, a private network linking only your own devices.
Do this instead: Reach the Gauntlet and the Hermes dashboards over Tailscale or a tunnel only.
Said by Julian Goldie, Alex Finn, Nimrod, Pav Rusovs
Why: Metics Media's baseline is a non-root sudo user, ed25519 SSH keys and a firewall allowing only SSH, 80 and 443. His Dokploy guide shows the Umami template shipping a default admin login of admin and umami, and he turns on weekly backups.
Do this instead: Change defaults on day one, keep keys-only SSH, close every port Coolify does not need, and switch backups on.
Said by Metics Media
Why: Tina Huang runs OpenClaw's security audit twice a day and her Hermes setup posts security reports to an alerts channel. NetworkChuck watches Hermes cron jobs, gateway and model logins with Checkmk. Wanderloots keeps the terminal in Docker so agent code stays boxed in.
Do this instead: Schedule a recurring audit that reports to Sentinel's channel, never with secrets in the message.
Said by Tina Huang, NetworkChuck, Wanderloots
Context and architecture mistakes: what the agent reads every time, where facts live, and who may write where.
Why: Eric Michaud splits his vault so agents may not edit the human side and the machine side is theirs. Wanderloots adds an allow-list so agents touch only approved vaults.
Do this instead: Mark human-only and machine-writable folders in an agents file at the vault root, and archive rather than delete.
Said by Eric Michaud, Wanderloots
Why: Sharbel A. says Hermes allows only one external provider at a time, among Honcho, Hindsight, Mem0 and Supermemory. Wanderloots gives his provider to the orchestrator alone. A catalogue note says Hindsight overlaps Chris's Qdrant plan.
Do this instead: Choose one memory layer per agent, and decide between a provider and Qdrant before building either.
Said by Sharbel A., Wanderloots
Why: Soul, memory and instruction files are read every session. Jack Roberts's Context Doctor audits Hermes for bloat in the soul file, memory, skill descriptions, tools and config. Pav Rusovs keeps CLAUDE.md to about 40 lines as a map.
Do this instead: Keep always-loaded files short, point to detail files, and send long reports to a documents area (Komputer Mechanic's Documents tab).
Said by Jack Roberts, Pav Rusovs, Komputer Mechanic
Why: Mason Fitzgerald calls CLAUDE.md a soft instruction that is not enforced. Simon Scrapes describes hooks as deterministic checks that run when Claude says it is finished.
Do this instead: Put must-never rules in permissions, hooks or scripts, and keep the instruction file for guidance.
Said by Mason Fitzgerald, Simon Scrapes
Why: Mason Fitzgerald claims retrieval was 5.2 times faster and 3.5 times cheaper across 15 questions, and his package sits behind a paid community. Pav Rusovs measured a plain map at 30 seconds against 62 and 17k tokens against 28k.
Do this instead: Run the same questions with and without Qdrant on Chris's own vault and compare time and tokens.
Said by Mason Fitzgerald, Pav Rusovs
Why: Jack Roberts recommends backing up the whole Hermes agent to GitHub daily. Wanderloots uses Git save points for the vault. Alex Finn suggests at least two profiles so one agent can repair the other.
Do this instead: Back up each agent's config, soul and memory files daily, and keep a second agent able to fix the first.
Said by Jack Roberts, Wanderloots, Alex Finn
Why: Tina Huang describes the LLM wiki as three jobs, ingest, query and lint, with an index and a log. Wanderloots uses scripts that check links and front matter, and Pav Rusovs runs a vault-lint skill.
Do this instead: Schedule a weekly lint job that reports broken links and bad front matter.
Said by Tina Huang, Wanderloots, Pav Rusovs
Reminders, schedules, your daily briefing and task tracking. 31 picks chosen for this lane. These are suggestions only. Nothing is installed until Sentinel has looked and you say yes.
Why Alfred: The main program Alfred runs on, with memory, skills, scheduled jobs and chat channels.
The main program that runs your always-on agents. It has memory, skills, scheduled jobs and over 20 chat channels.
Suits: Alfred, Brainiac, Cerebra. It is the runtime all seven of Chris's agents already use.
In your videos: Alex Finn calls it the greatest AI tool ever made · Named in 56 of your videos
Popularity: 253.6k stars
Watch it here:
and 52 more videos
Why Alfred: Runs the 07:00 briefing and other timed tasks by itself and posts to a chat channel.
Runs a task on a timer by itself and sends the result to a chat channel.
Suits: Brainiac. Brainiac runs Chris's cron work.
In your videos: Alex Finn says to prune and pause old cron jobs to improve speed and cost · Named in 13 of your videos
Watch it here:
and 9 more videos
Why Alfred: Gives Alfred a task board to track who holds each request and its status.
A task board where work moves from new to done and gets shared out between helper agents.
Suits: Forge, Brainiac. Forge runs builds and ops; Brainiac could own filing tasks on the board.
In your videos: Named in 10 of your videos
Watch it here:
and 6 more videos
Why Alfred: Writes a short note of a session so another agent can pick up the work.
Writes a short handoff note of a session so another agent can carry on.
Suits: Alfred, Forge. Alfred and Forge both hand work between agents and surfaces.
In your videos: Named in 3 of your videos
Watch it here:
Why Alfred: A page for Chief to see sessions, jobs and logs behind the daily routines.
A web page to see and manage your Hermes sessions, models, jobs, skills, keys, channels and logs.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Why Alfred: Runs plain timed jobs as a script, saving tokens on routine reminders and weather pulls.
A scheduled-job option that runs only a script, saving tokens by skipping the AI.
Suits: Brainiac. Brainiac's scheduled vault jobs could run as plain scripts.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Lets scheduled jobs remember earlier runs, so briefings do not repeat the same items.
Lets scheduled jobs remember earlier runs so they do not repeat work.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Lets Chief nudge Alfred mid-task without stopping the work.
Lets you nudge the agent while it works without stopping it.
Suits: Alfred, Forge. Alfred, the general assistant, could use it in everyday chats; Forge could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Gives Chief a quick way to end a task or background process that has gone wrong.
Stops the current task or ends background processes.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Why Alfred: Lets him run a second task, such as a weather check, while another is going.
Runs a second task in the background at the same time.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Why Alfred: Lets Chief hand Alfred tasks by speaking, which suits a busy day.
Talk to Hermes out loud and use a wake word to hand it tasks.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Alex Finn says ChatGPT voice mode needs an OpenAI API key, about 5 cents a minute
In your videos: Named in 3 of your videos
Watch it here:
Why Alfred: Matches his 07:00 hrs briefing: a daily job that briefs Chief on the top priority.
A daily scheduled job that asks your top priority, briefs you and starts work.
Suits: Alfred, Brainiac. Alfred, the general assistant, could use it in everyday chats; Brainiac could also use it.
In your videos: Named in 4 of your videos
Watch it here:
Why Alfred: Gives each briefing item and task card one clear sentence saying what finished looks like.
One measurable sentence saying what done looks like for each agent task.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Fits his 22:45 hrs diary: record what was done before replying, in one place.
Each agent writes down what it did, or adds it to today's daily note, before replying.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 3 of your videos
Watch it here:
Why Alfred: Lets him write replies for Chief to approve, and never send anything himself.
The agent writes email replies in your voice but never sends them on its own.
Suits: Alfred, Sentinel. Alfred, the general assistant, could use it in everyday chats; Sentinel could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Gives a simple way to carry open items into a fresh chat without losing them.
Ask for a handoff file at around 200 to 300k tokens, then restart in a fresh chat.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: A model for showing tasks, agent status and notes on one page for Chief.
A custom dashboard your agent builds to show tasks, content, memory, docs and agent status.
Suits: Cyborg. Cyborg runs the LAN dashboard Chris is planning.
In your videos: Named in 4 of your videos
Watch it here:
Why Alfred: Turns action items from a call into tasks that he can log and hand out.
After each call, Hermes pulls the transcript and turns action items into tasks.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: An inbox-then-file routine that fits his intake, with Brainiac doing the filing.
An hourly routine gathers Slack, meetings and email into an inbox, then a librarian step files it.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Chief can send a voice note that Alfred turns into a reminder or task.
Send voice notes from your phone that an agent on a home machine turns into text and acts on.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Why Alfred: Blocks sending, sharing and deleting on the server, matching his never-send, never-delete rules.
A settings file that blocks sending, sharing and deleting on the server. Email stays draft only.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Helps him measure and trim his own usage on repeated daily jobs.
Have the agent measure and cut its own usage on each task.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Matches the vault Inbox rule: drop notes in an inbox and let a script file them.
Instructions tell agents to read the wiki first and drop notes into an inbox that a script files.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: A simple test for which repetitive tasks are worth handing to Alfred.
Pick tasks that are repetitive, rules-based and give time back.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Lets daily notes hold numbers in headers, useful for tracking status over time.
Scheduled scripts write numbers into daily-note headers so your notes work like a table you can query.
Suits: Brainiac, Cyborg. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cyborg could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Why Alfred: Breaks a larger request into task cards, which suits his task intake and dispatch work.
Breaks a product requirements document into tasks, using 36 MCP tools.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Free model for small side jobs like weather lines and short summaries, saving budget.
Nvidia open models, free on OpenRouter for small side jobs your agents do.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 4 of your videos
Watch it here:
Why Alfred: Offers a quick-capture inbox and focus view that suits reminders and daily priorities.
Custom Obsidian plugin with focus, metrics, a heat map and a quick-capture inbox.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
Watch out: Sold through his community
In your videos: Named in 1 of your videos
Watch it here:
Why Alfred: Shows tasks and health across all the agents, helping him keep the Task-Tracker honest.
A dashboard for several Hermes agents showing tasks, health checks and live costs.
Suits: Cyborg, Forge. Cyborg runs the dashboard side and Forge runs fleet ops; matches Chris's planned agent dashboard.
In your videos: Sharbel presents it as his own tool for running a Hermes fleet · Named in 2 of your videos
Watch it here:
Why Alfred: Makes briefings and summaries for Chief read like a person wrote them, not a machine.
Rewrites text to remove the usual signs that an AI wrote it.
Suits: Alfred, Mantis. Alfred and Mantis write text for Chris; it is already built into Hermes.
Watch out: Teku AI says it is token heavy; Sharbel notes the skill file is 29.6 KB
In your videos: Playlist-1 doc notes it is already built into Hermes · Named in 4 of your videos
Popularity: 60.1k stars
Watch it here:
Why Alfred: Keeps plan and progress in files, so multi-step requests are not lost between chats.
Keeps plan, findings and progress files so the agent does not lose its plan.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 27.5k stars
Watch it here:
Keeps the vault tidy: filing, links, tags and search. 32 picks chosen for this lane. These are suggestions only. Nothing is installed until Sentinel has looked and you say yes.
Why Brainiac: The vault itself; plain markdown files she reads, files into and links every day.
Note app using plain markdown files. Your agents use it as their memory, wiki and workspace.
Suits: Brainiac, Cyborg, Cerebra. Brainiac files the vault, Cyborg publishes it and Cerebra indexes it.
Watch out: Mason Fitzgerald argues file search over Obsidian is slow and costly versus RAG
In your videos: Named in 20 of your videos
Watch it here:
and 16 more videos
Why Brainiac: Plain markdown notes joined by wiki links, close to how her vault is built.
Agent memory kept as plain markdown files with wiki links. It replaces Notion.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
Watch out: Hosted plan costs $15 a month
In your videos: Named in 1 of your videos
Popularity: 3.6k stars
Watch it here:
Why Brainiac: Lets her work on the vault directly for searching, moving and linking notes.
An Obsidian setting that lets an agent work on your vault directly.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Short handoff notes between agents, which she receives and files from Inbox.
Writes a short handoff note of a session so another agent can carry on.
Suits: Alfred, Forge. Alfred and Forge both hand work between agents and surfaces.
In your videos: Named in 3 of your videos
Watch it here:
Why Brainiac: Finds past conversations, helping her check what was already filed or decided.
Searches your old past conversations with the agent so you can find what was said.
Suits: Alfred, Brainiac. Alfred and Brainiac can recall earlier chats on demand.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Why Brainiac: Her wiki job follows this pattern: ingest sources, link pages, and lint the result regularly.
A pattern where an LLM turns raw sources into a linked markdown wiki, with ingest, query and lint steps.
Suits: Brainiac, Cerebra. Brainiac maintains the vault and Cerebra indexes it; this is the method behind both.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Why Brainiac: Same flow as her Inbox rule: read the wiki first, then file new notes.
Instructions tell agents to read the wiki first and drop notes into an inbox that a script files.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: A searchable wiki of decisions and logs, kept current, which is what Wiki/index.md is.
A searchable wiki of decisions, logs and research that your agent keeps up to date so it stays consistent.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 4 of your videos
Watch it here:
Why Brainiac: Supports her rule of not editing other agents' or Chris's notes beyond approved places.
A rule that agents may not edit the human side of your vault. The machine side is theirs.
Suits: Brainiac, Sentinel. Brainiac and Sentinel already guard off-limits folders like the archive.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Fits her habit of writing one clean log line for every change she makes.
Each agent writes down what it did, or adds it to today's daily note, before replying.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 3 of your videos
Watch it here:
Why Brainiac: Gives a way back after each filing change, without anything ever being deleted.
Saves versions of your agent or vault, with .gitignore, and backs it up daily to GitHub.
Suits: Cyborg. Cyborg already publishes the vault to GitHub.
In your videos: Named in 3 of your videos
Watch it here:
Why Brainiac: Index files and a short map help agents find notes fast, like 00-Index.
A short CLAUDE.md map plus index files, built from an LLM interview, so agents find context fast.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Pav Rusovs measured 30s vs 62s and 17k vs 28k tokens with the map · Named in 1 of your videos
Watch it here:
Why Brainiac: Folders of markdown with README files, the same shape as her index pages.
Folders of markdown notes with README files that agents search, read and write back to.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Mirrors her work: items gather in an inbox, then a librarian step files them.
An hourly routine gathers Slack, meetings and email into an inbox, then a librarian step files it.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Turns kept decisions and working notes into lessons she can link and index.
Keep decisions and working docs, then turn them into lessons in the brain.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Daily note headers hold data, helpful for tracking the nightly note she checks.
Scheduled scripts write numbers into daily-note headers so your notes work like a table you can query.
Suits: Brainiac, Cyborg. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cyborg could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Why Brainiac: Long reports are saved as documents, not chat, which suits her filing intake.
Agents save long reports in a documents area, not chat, which saves tokens.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Exports what a system knows as markdown, a simple note she could file and link.
Export what your AI system knows about you as markdown, to set up a new agent.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Database-style views over notes, useful for index pages and tag-based overviews.
Database-style views over your vault notes.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Brings clipped pages into the vault, which land in Inbox for her to file.
Saves web pages and YouTube transcripts into your vault's raw folder.
Suits: Cerebra, Brainiac. Cerebra does web and source-graded research and vault indexing, which this feeds; Brainiac could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: A visual board for mapping how notes and folders link together.
A visual board in Obsidian for charts, boards and views.
Suits: Brainiac, Mantis. Brainiac files and maintains the vault and runs cron jobs, which this supports; Mantis could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Why Brainiac: Surfaces related notes in the vault, useful for spotting links and orphans.
Community Obsidian plugin that links AI to your vault.
Suits: Brainiac, Cerebra. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cerebra could also use it.
In your videos: Named in 1 of your videos
Popularity: 5.5k stars
Watch it here:
Why Brainiac: Ready-made ingest, query, maintain and lint steps plus note templates to borrow for the wiki.
Starter kit with a wiki_tool.py script, ingest, query, maintain and lint skills, and note templates.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
Watch out: Starter kit is for members only
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Long-term memory that can retain and recall, relevant to keeping a vault consistent.
Long-term agent memory that can retain, recall and reflect. A Hermes memory option that runs in Docker.
Suits: Brainiac, Cerebra. Brainiac files and maintains the vault and runs cron jobs, which this supports; Cerebra could also use it.
Watch out: Playlist-1 doc notes it overlaps Chris's Qdrant plan
In your videos: Named in 4 of your videos
Popularity: 44.6k stars
Watch it here:
Why Brainiac: Matches her role: a librarian files reports into the wiki and asks Chris to approve reviews.
Hermes skill where a librarian bot files reports into your Obsidian wiki and asks you to approve reviews.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 2 of your videos
Watch it here:
Why Brainiac: Checks for broken links and rebuilds the map, helping her keep zero orphan notes.
Pav Rusovs' skills: nightly map rebuild, broken-link check, Mac cleanup and a media skill.
Suits: Brainiac. Brainiac maintains the vault and could lint broken links nightly.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Teaches agents correct Obsidian markdown, Bases and links, so her notes stay clean.
Skill pack from Obsidian's CEO that teaches agents the Obsidian command line, markdown, Bases, Canvas and Defuddle.
Suits: Brainiac. Brainiac writes vault notes and needs correct wikilinks and front matter.
In your videos: Named in 2 of your videos
Popularity: 50.0k stars
Watch it here:
Why Brainiac: Limits which vaults an agent may touch, matching her off-limits folders and archive rules.
Script and skill that limit which vaults an agent is allowed to touch.
Suits: Sentinel, Brainiac. Sentinel and Brainiac could enforce the archive-folder ban with an allow-list.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Her lane already uses it on index descriptions and summaries written for Chris.
Rewrites text to remove the usual signs that an AI wrote it.
Suits: Alfred, Mantis. Alfred and Mantis write text for Chris; it is already built into Hermes.
Watch out: Teku AI says it is token heavy; Sharbel notes the skill file is 29.6 KB
In your videos: Playlist-1 doc notes it is already built into Hermes · Named in 4 of your videos
Popularity: 60.1k stars
Watch it here:
Why Brainiac: Built-in skill to save and read vault notes, covering her everyday filing work.
A built-in skill that saves and reads notes in your Obsidian vault as extra memory.
Suits: Brainiac. Brainiac files notes into the vault.
In your videos: Named in 1 of your videos
Watch it here:
Why Brainiac: Cleans web pages into plain text so clips are easier to read and file.
Strips a web page down to clean readable text, which saves on tokens.
Suits: Cerebra, Brainiac. Cerebra fetches pages for research and Brainiac clips into the vault; both save tokens.
In your videos: Named in 4 of your videos
Popularity: 9.4k stars
Watch it here:
Why Brainiac: Spots clutter in files and memory and sets items aside instead of deleting them.
Checks Hermes for clutter in its files, memory and tools, and sets items aside instead of deleting.
Suits: Forge, Sentinel. Forge runs agent ops and Sentinel reviews skills, so both benefit from a bloat audit that archives instead of deleting.
In your videos: Featured as the headline skill; the playlist-1 doc made it a top pick · Named in 1 of your videos
Watch it here:
Web research and checking sources, public pages only. 30 picks chosen for this lane. These are suggestions only. Nothing is installed until Sentinel has looked and you say yes.
Why Cerebra: Plain markdown notes where his research notes live and link to Cooper project pages.
Note app using plain markdown files. Your agents use it as their memory, wiki and workspace.
Suits: Brainiac, Cyborg, Cerebra. Brainiac files the vault, Cyborg publishes it and Cerebra indexes it.
Watch out: Mason Fitzgerald argues file search over Obsidian is slow and costly versus RAG
In your videos: Named in 20 of your videos
Watch it here:
and 16 more videos
Why Cerebra: Answers only from sources he uploads, handy for checking claims against saved records.
A notebook that answers from your own sources, and makes audio, video, slides, mind maps and reports.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Why Cerebra: Stores scanned records and documents with tags and search, suited to family-history paperwork.
Stores your documents, adds tags by rules, and lets you search them.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: Local workspace for uploading documents and asking questions, good for Cooper source bundles.
Local workspace where you upload documents and run agent workflows.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: One key for many models with spending limits, plus free models for low-cost research drafts.
One API key that reaches hundreds of models, with spending limits per key, rankings and free models.
Suits: Forge. Forge manages models and each agent's OpenRouter key and budget.
In your videos: Moe Lueker uses it as a cost router to the cheapest capable model · Named in 14 of your videos
Watch it here:
and 10 more videos
Why Cerebra: The program that runs his, with memory, skills and scheduled jobs for steady research work.
The main program that runs your always-on agents. It has memory, skills, scheduled jobs and over 20 chat channels.
Suits: Alfred, Brainiac, Cerebra. It is the runtime all seven of Chris's agents already use.
In your videos: Alex Finn calls it the greatest AI tool ever made · Named in 56 of your videos
Popularity: 253.6k stars
Watch it here:
and 52 more videos
Why Cerebra: Runs multi-agent research that cross-checks findings, helpful for confirming or contradicting earlier Cooper claims.
Multi-agent research inside Claude Code that cross-checks its findings.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: Searches past conversations, so he can recall earlier leads without repeating paid searches.
Searches your old past conversations with the agent so you can find what was said.
Suits: Alfred, Brainiac. Alfred and Brainiac can recall earlier chats on demand.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Why Cerebra: Switch-on connections inside Hermes, so he can add research tools one at a time.
Switch-on connections inside Hermes to tools like Canva, Notion, GitHub and NotebookLM.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Why Cerebra: Runs his browsing inside Docker, keeping hostile web pages walled off from the server.
Runs agent code and browsing inside Docker so it stays walled off from the rest of your server.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Why Cerebra: Turns tools like browser automation on or off per platform, keeping his access to what the job needs.
Turns tools such as browser automation or computer use on or off for each platform.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Why Cerebra: Turns raw sources into a linked markdown wiki, a good shape for George Cooper research notes.
A pattern where an LLM turns raw sources into a linked markdown wiki, with ingest, query and lint steps.
Suits: Brainiac, Cerebra. Brainiac maintains the vault and Cerebra indexes it; this is the method behind both.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Why Cerebra: Keeps a searchable record of decisions and research, so his findings stay consistent over time.
A searchable wiki of decisions, logs and research that your agent keeps up to date so it stays consistent.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 4 of your videos
Watch it here:
Why Cerebra: Folders of markdown with README files he can search and write back to, matching his vault work.
Folders of markdown notes with README files that agents search, read and write back to.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: Keeps long research write-ups in a documents area instead of chat, saving tokens.
Agents save long reports in a documents area, not chat, which saves tokens.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: The standard way his tools connect, which helps his understand and check each tool added.
The standard way agents and servers share tools, resources and prompts. It is set up in mcp.json.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Why Cerebra: Opens and reads public pages through a browser, helping with sites that need scripts to display.
Microsoft's official tool for controlling a browser using page accessibility snapshots.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Popularity: 38.0k stars
Watch it here:
Why Cerebra: Lets him put gathered sources into NotebookLM and ask questions grounded only in them.
Connects Hermes to NotebookLM for research and a podcast-style daily brief.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: Searches the web as it answers, giving a quick cited lead to open and verify.
A model that searches the web as it answers. Used by agents for web research.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 3 of your videos
Watch it here:
Why Cerebra: Saves public web pages into the vault as clean notes, keeping a copy of each source.
Saves web pages and YouTube transcripts into your vault's raw folder.
Suits: Cerebra, Brainiac. Cerebra does web and source-graded research and vault indexing, which this feeds; Brainiac could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: Researches a topic on its own and writes a sourced report he can then verify by hand.
An agent that researches a topic on its own and writes a report.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Popularity: 29.5k stars
Watch it here:
Why Cerebra: Open-source scraper that turns public pages into clean text he can read and cite.
An open-source web scraper, the second choice after the main one.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: Chats with his own PDFs on the local machine, so archive scans stay private.
Chat with your own PDFs on your computer, at localhost:8001.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: Reads public pages on X, Reddit, YouTube and GitHub without API fees, matching his free-first search order.
Lets agents read sites like X, Reddit, YouTube and GitHub without paying for API fees.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Teku AI cites about 81k stars and says it gets past 403 bot blocks · Named in 4 of your videos
Popularity: 101.3k stars
Watch it here:
Why Cerebra: Adds footnote-style sources to his findings, so every claim shows the page it came from.
Adds footnote-style sources to research results so you can see where facts came from.
Suits: Cerebra. Cerebra's source-graded research needs citations.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: Gives him a multi-step research routine for bigger questions, with several models cross-checking each other.
A multi-step research process that uses several models at once.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 2 of your videos
Watch it here:
Why Cerebra: Gathers recent public discussion on a topic from several sites, useful for current source leads.
Researches a topic from the last 30 days across Reddit, X, YouTube, GitHub and the web.
Suits: Cerebra. Cerebra does web research and this gives her fresh multi-platform sentiment.
In your videos: AI Edge notes 140 contributors; Futurepedia lists it first · Named in 3 of your videos
Popularity: 65.2k stars
Watch it here:
Why Cerebra: Official skills to read PDF, Word and Excel files, which many census and record downloads use.
Official Anthropic skills for working with PDF, Excel, PowerPoint and Word files.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Why Cerebra: Fetches video transcripts, channels and search with no API key, for lectures and archive talks.
Replaces the default YouTube skill with transcripts, channels and search, with no API key.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Why Cerebra: Pulls a video transcript he can read and quote, instead of trusting a summary.
A bundled skill that gets a video transcript for summaries or scheduled quizzes.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 1 of your videos
Watch it here:
Apps and code, cyber design, agentic AI, sim racing, markets. 39 picks chosen for this lane. These are suggestions only. Nothing is installed until Sentinel has looked and you say yes.
Why Cyborg: The program that runs his always-on agent loops, memory, skills and scheduled jobs.
The main program that runs your always-on agents. It has memory, skills, scheduled jobs and over 20 chat channels.
Suits: Alfred, Brainiac, Cerebra. It is the runtime all seven of Chris's agents already use.
In your videos: Alex Finn calls it the greatest AI tool ever made · Named in 56 of your videos
Popularity: 253.6k stars
Watch it here:
and 52 more videos
Why Cyborg: Private encrypted network linking his agent to the PS5 telemetry and Chris's devices.
A private, encrypted network so your agents can reach and log into all your devices.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 8 of your videos
Popularity: 36.3k stars
Watch it here:
and 4 more videos
Why Cyborg: Makes a local telemetry app reachable safely without opening ports.
Makes a local model or app reachable from the internet.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Named in 1 of your videos
Watch it here:
Why Cyborg: Gives his self-hosted dashboards HTTPS behind a proper front door.
A front door that gives the Hermes dashboard HTTPS in Hostinger's Docker template.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Why Cyborg: One key reaching many models, with spend limits per key for his orchestration tests.
One API key that reaches hundreds of models, with spending limits per key, rankings and free models.
Suits: Forge. Forge manages models and each agent's OpenRouter key and budget.
In your videos: Moe Lueker uses it as a cost router to the cheapest capable model · Named in 14 of your videos
Watch it here:
and 10 more videos
Why Cyborg: Catalogues to browse for his skill scouting, then pass to Sentinel.
Catalogues of shared agent skills you can browse.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
Watch out: Tina Huang avoids ClawHub skills unless from trusted authors
In your videos: Named in 2 of your videos
Watch it here:
Why Cyborg: Main terminal coding agent for his Swift, full-stack and telemetry builds, with skills, hooks and MCP.
Anthropic's coding agent for the terminal, with skills, plugins, hooks, MCP and helper agents.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Pav Rusovs chose it over Hermes, citing Anthropic rules on subscriptions in third-party harnesses · Named in 39 of your videos
Popularity: 153.3k stars
Watch it here:
and 35 more videos
Why Cyborg: Tests the MCP servers he builds before anything depends on them.
Tests MCP servers. Run it with npx @modelcontextprotocol/inspector.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 10.7k stars
Watch it here:
Why Cyborg: Packages telemetry services and test apps so they run separately and cleanly.
Runs several apps and agents separately on one machine. Docker Compose handles whole stacks.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Why Cyborg: Tests the web front ends he builds in a real browser.
A Microsoft tool that opens a real browser to test your app. Lighter than the MCP version.
Suits: Cyborg, Forge. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches; Forge could also use it.
In your videos: Named in 1 of your videos
Popularity: 98.3k stars
Watch it here:
Why Cyborg: Makes managing those Git worktrees easy when running parallel coding agents.
Manages Git worktrees, which suits running several Claude Code agents at the same time.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Why Cyborg: Runs local models for private analysis and cheap background loops.
Manages local models and serves them at localhost:11434. Works with Hermes and Ollama Cloud.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 9 of your videos
Popularity: 184.7k stars
Watch it here:
and 5 more videos
Why Cyborg: Lets him hand narrow jobs to small helpers and check their work.
Small helper agents that the main agent starts for narrow jobs, then checks their work.
Suits: Forge, Cerebra. Forge and Cerebra can split work into parallel helpers.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Why Cyborg: Keeps him working until a stated goal is done, with a turn limit.
Keeps the agent working until your stated goal is done, with a limit on turns.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Jack Roberts says a goal has a 20-turn budget by default
In your videos: Sharbel says to have Hermes write the goal prompt first (meta-prompting) · Named in 5 of your videos
Watch it here:
and 1 more videos
Why Cyborg: Runs plain scripts on schedule without model cost, ideal for the telemetry and market daemons.
A scheduled-job option that runs only a script, saving tokens by skipping the AI.
Suits: Brainiac. Brainiac's scheduled vault jobs could run as plain scripts.
In your videos: Named in 1 of your videos
Watch it here:
Why Cyborg: Makes him ask Chris first before spending, sending or deleting, matching his lane rules.
Makes the agent ask you first before it spends money, sends things or deletes anything.
Suits: Sentinel, Forge. Sentinel enforces safety and Forge runs ops; gates match Chris's no-surprise rules.
In your videos: Named in 2 of your videos
Watch it here:
Why Cyborg: Runs his experimental code walled off from the rest of the server.
Runs agent code and browsing inside Docker so it stays walled off from the rest of your server.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Why Cyborg: Small checks that run on events, keeping his code work tidy and safe.
Small scripts that run on events in Claude Code to check work or add context.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 3 of your videos
Watch it here:
Why Cyborg: Finds past conversations so he can recall earlier tuning and design decisions.
Searches your old past conversations with the agent so you can find what was said.
Suits: Alfred, Brainiac. Alfred and Brainiac can recall earlier chats on demand.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Why Cyborg: Keeps his background loops running when one model provider fails.
Backup models and accounts so your agent keeps working when one provider fails.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Why Cyborg: Matches his LLM orchestration work: one strong planner, cheap models doing narrow jobs.
One strong model plans and hands work to cheap models, each given only the tools it needs.
Suits: Forge. Forge sets models; the doc notes Chris's agents all run Haiku, so this is the part not yet used.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Why Cyborg: Keeps his market and data access read-only, limiting damage if something goes wrong.
Give agents read-only keys and database users that cannot see tables holding personal data.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Cyborg: Reminds him to store exchange and API keys as environment variables, never in chat.
Never paste keys into chat. Set them as environment variables in the terminal and give least access.
Suits: Sentinel. Sentinel enforces Chris's no-secrets rule.
In your videos: Named in 1 of your videos
Watch it here:
Why Cyborg: Guides secure network design for dashboards, matching his cyber defence architecture work.
Open the Hermes dashboard only through a VPN or SSH tunnel, never openly on the internet.
Suits: Sentinel, Cyborg. Sentinel reviews and security-vets skills, which this supports; Cyborg could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Cyborg: Lets him work on several branches of one app at once.
Separate git worktrees let several agents work on one repo at the same time.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Why Cyborg: A pattern for his own live view of telemetry, markets and agent status.
A custom dashboard your agent builds to show tasks, content, memory, docs and agent status.
Suits: Cyborg. Cyborg runs the LAN dashboard Chris is planning.
In your videos: Named in 4 of your videos
Watch it here:
Why Cyborg: Lets him save repeated tuning and research routines as reusable skills.
Your agent saves repeated jobs, checklists or SOPs as its own reusable skills.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Why Cyborg: Fits his daily diary habit by recording what he did each day.
Each agent writes down what it did, or adds it to today's daily note, before replying.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 3 of your videos
Watch it here:
Why Cyborg: A searchable record of his decisions and research that keeps his work consistent.
A searchable wiki of decisions, logs and research that your agent keeps up to date so it stays consistent.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 4 of your videos
Watch it here:
Why Cyborg: Feeds him up-to-date library documentation while coding Swift, API and web pipelines.
Feeds the agent current documentation for code libraries.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 63.7k stars
Watch it here:
Why Cyborg: Keeps a lasting map of his code so he spends fewer tokens exploring.
Turns your code into a lasting knowledge graph so the agent uses fewer tokens exploring.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Claims 99% (Sharbel) or 120x (Teku AI) fewer tokens; about 45k stars · Named in 4 of your videos
Popularity: 47.2k stars
Watch it here:
Why Cyborg: Turns forums and sites into clean text for his market and sentiment watching.
Turns websites into clean markdown. It can scrape, crawl and get past bot detection.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
Watch out: The Next New Thing offers Scrapling as a free replacement
In your videos: Named in 4 of your videos
Popularity: 194.9k stars
Watch it here:
Why Cyborg: Cheap, fast model for small helper tasks and background scouting loops.
Anthropic's fast, cheap model. Good for small coding helper agents where cost matters.
Suits: Alfred, Brainiac. All seven agents run on Haiku today.
In your videos: Sharbel suggests Haiku sub-agents to protect the main cache · Named in 8 of your videos
Watch it here:
and 4 more videos
Why Cyborg: Local coding model he can run on his own machine for private code help.
Coding models you can run on your own machine.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Why Cyborg: Python kit for building agent teams, useful for his agentic orchestration experiments.
Python kit for building teams of agents, shown working with Ollama models.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Why Cyborg: Lets him build his own MCP servers, such as one feeding telemetry or market data.
Python kit for building your own MCP servers and clients.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 24.3k stars
Watch it here:
Why Cyborg: Shows his results as visual screens like dashboards, useful for telemetry and market views.
Makes agent replies appear as visual screens like dashboards and forms.
Suits: Cyborg. Cyborg publishes the vault to GitHub and runs the LAN dashboard, which this touches.
In your videos: Claimed 13,000 stars · Named in 1 of your videos
Watch it here:
Why Cyborg: Gives him current Swift and SwiftUI documentation when building the iOS telemetry app.
Gives agents current Swift and SwiftUI documentation for building Apple apps.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Why Cyborg: Researches the last month across Reddit, X, GitHub and the web, fitting market watching.
Researches a topic from the last 30 days across Reddit, X, YouTube, GitHub and the web.
Suits: Cerebra. Cerebra does web research and this gives her fresh multi-platform sentiment.
In your videos: AI Edge notes 140 contributors; Futurepedia lists it first · Named in 3 of your videos
Popularity: 65.2k stars
Watch it here:
Editing, writing, photo, video and PDF work. 40 picks chosen for this lane. These are suggestions only. Nothing is installed until Sentinel has looked and you say yes.
Why Forge: Free Adobe-style apps with built-in MCP, the base for her photo, video and PDF editing.
Free Adobe-style apps built in Rust, with a built-in MCP so agents can run them.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Mostly early alpha
In your videos: Named in 4 of your videos
Watch it here:
Why Forge: Shrinks and converts files on the Mac, handy for resizing photos and PDFs before return.
A Mac app that shrinks and converts files on your clipboard. Has an MCP for Claude.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Turns a prompt into an editable animation that Chief can adjust with comments.
Turns a prompt into an editable animation that you tweak with comments.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Beta for Team and Enterprise plans only
In your videos: Named in 2 of your videos
Watch it here:
Why Forge: A design editor with an MCP, useful for layout touch-ups on photos and documents.
A design editor, also available as an MCP in Hermes.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Why Forge: Design canvas for mocking up drafts, so Chief can see a draft before changes are built.
Claude's design canvas for mocking up apps and screens with colour options.
Suits: Mantis. Mantis does visual work, and Chris wants drafts made in Claude Design.
In your videos: Named in 2 of your videos
Watch it here:
Why Forge: The vault where drafts live, and where Forge saves edited copies in her own folder.
Note app using plain markdown files. Your agents use it as their memory, wiki and workspace.
Suits: Brainiac, Cyborg, Cerebra. Brainiac files the vault, Cyborg publishes it and Cerebra indexes it.
Watch out: Mason Fitzgerald argues file search over Obsidian is slow and costly versus RAG
In your videos: Named in 20 of your videos
Watch it here:
and 16 more videos
Why Forge: Answers from your own sources, helpful for checking a draft against its source notes.
A notebook that answers from your own sources, and makes audio, video, slides, mind maps and reports.
Suits: Cerebra. Cerebra does web and source-graded research and vault indexing, which this feeds.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Why Forge: A teleprompter showing a script, handy for checking how a written draft reads aloud.
A teleprompter on your Mac that shows your script while you read it.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Local image generation and editing on Mac, keeping photos on Chief's own machine.
Fast local image generation on Apple silicon Macs and iPhone.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Why Forge: A file telling the agent who it is, keeping her four edit levels and limits steady.
A file that tells each agent who it is and how to behave.
Suits: Forge. Forge maintains the agents' own files.
In your videos: Jack Roberts calls it a character bible and gives a template · Named in 10 of your videos
Watch it here:
and 6 more videos
Why Forge: Blocks edits until a plan is approved, which suits her careful restructure level.
Blocks edits until you approve a plan.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
Watch out: Simon Scrapes says the Claude Code team is considering removing it
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Turns a PDF or link into a saved skill, useful for building her approved craft skills.
Turns a PDF, link or YouTube video into a new saved skill for your agent.
Suits: Brainiac, Cerebra. Brainiac and Cerebra could turn sources into reusable skills.
In your videos: Named in 2 of your videos
Watch it here:
Why Forge: A voice guide with banned words helps Forge keep Chief's tone during every edit.
A doc built by interview about how you think, plus a voice guide per medium with banned words.
Suits: Alfred, Mantis. Alfred, the general assistant, could use it in everyday chats; Mantis could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Matches her rule of never editing originals; she only writes copies in her folder.
A rule that agents may not edit the human side of your vault. The machine side is theirs.
Suits: Brainiac, Sentinel. Brainiac and Sentinel already guard off-limits folders like the archive.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Chains small skills in order, like proofread then copyedit then clarity then restructure.
One big skill runs smaller skills in order, such as microsite, then copy, then QA.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: One shared markdown skill folder lets Forge read the same editing methods each time.
A team folder of markdown skill files that every agent reads.
Suits: Forge, Brainiac. Forge builds and runs agent operations, where this belongs; Brainiac could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: A lessons file Forge reads before each edit run, keeping her style habits fresh.
Each workflow folder holds lessons learned, which the agent reads before every run.
Suits: Brainiac, Forge. Brainiac files and maintains the vault and runs cron jobs, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Blocks sending and deleting, with email draft only, matching her no-publish rule.
A settings file that blocks sending, sharing and deleting on the server. Email stays draft only.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Notes with README files that agents search, giving Forge style context for edits.
Folders of markdown notes with README files that agents search, read and write back to.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Image and video editing through a connector, with face-consistent results for photo work.
Makes images and video with major models through an MCP connector and command line, including face-consistent Soul.
Suits: Mantis. Mantis makes images and could generate them directly.
In your videos: Tristen O'Brien adds it to Claude as a custom connector · Named in 6 of your videos
Watch it here:
and 2 more videos
Why Forge: Lets Forge edit inside a picture, which fits photo editing on copies.
An image model that lets you edit inside the picture. In public beta.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Google image model that can alter and touch up photos for the photo lane.
Google's image model, used to make a daily post image.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 2 of your videos
Watch it here:
Why Forge: Image model for edits and generation, offered as a plugin and a Codex skill.
OpenAI's image maker, also usable through a Hermes plugin and a Codex skill.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
Why Forge: Makes editable SVG images that stay sharp at any size.
Makes editable SVG images, which are sharp at any size.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Makes video from prompts, a possible tool for the video editing lane.
ByteDance's model for making video from prompts.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 3 of your videos
Watch it here:
Why Forge: Controls DaVinci Resolve for real video editing, matching the filmcraft lane.
Lets Hermes control DaVinci Resolve Studio live for video editing.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Pauses risky changes and waits for approval, matching her never-overwrite and never-delete rules.
Pauses risky deletes, lists the affected files and waits for you to press Proceed or Cancel.
Suits: Sentinel. Sentinel would like a hard stop on deletes; it fits Chris's archive-not-delete rule.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Steps through each edit just made, so Chief can review exactly what changed.
Sample add-on with a replay button that steps through each edit Claude just made.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Removes AI-sounding tells from text, matching her copyedit level for notes and drafts, never code.
Rewrites text to remove the usual signs that an AI wrote it.
Suits: Alfred, Mantis. Alfred and Mantis write text for Chris; it is already built into Hermes.
Watch out: Teku AI says it is token heavy; Sharbel notes the skill file is 29.6 KB
In your videos: Playlist-1 doc notes it is already built into Hermes · Named in 4 of your videos
Popularity: 60.1k stars
Watch it here:
Why Forge: Official skills for PDF and Word files, useful for editing documents and returning separate copies.
Official Anthropic skills for working with PDF, Excel, PowerPoint and Word files.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Why Forge: Helps write and revise documents together with Chief, fitting her writing and clarity rewrite work.
Helps Claude write documents together with you.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Why Forge: Saves voice and visual rules so edited output stays consistent across drafts.
Saves a brand voice and visual rules so output stays consistent.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Why Forge: Lets Forge watch a video from a file, useful when reviewing footage for editing.
Lets an agent watch a video from a link or file.
Suits: Cerebra, Mantis. Cerebra does web and source-graded research and vault indexing, which this feeds; Mantis could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Writes an HTML page that turns into an MP4, a way to produce video edits.
The agent writes an HTML page that turns into an MP4 video.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
Watch out: Needs Node 22
In your videos: Named in 7 of your videos
Popularity: 71.8k stars
Watch it here:
and 3 more videos
Why Forge: Makes video from code, a route for filmcraft editing work and cuts.
Open-source tool that makes video from React code, with a map-animation skill.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Tristen O'Brien calls it the free-tier alternative to Claude Motion · Named in 7 of your videos
Popularity: 67.4k stars
Watch it here:
and 3 more videos
Why Forge: Makes graphics and saves them as PNG or PDF, handy for PDF output copies.
Makes social media graphics and saves them as PNG or PDF.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Popularity: 183.3k stars
Watch it here:
Why Forge: Teaches markdown and vault formats so edited copies keep notes intact.
Skill pack from Obsidian's CEO that teaches agents the Obsidian command line, markdown, Bases, Canvas and Defuddle.
Suits: Brainiac. Brainiac writes vault notes and needs correct wikilinks and front matter.
In your videos: Named in 2 of your videos
Popularity: 50.0k stars
Watch it here:
Why Forge: Plan and progress files stop Forge losing her place in long restructure edits.
Keeps plan, findings and progress files so the agent does not lose its plan.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 27.5k stars
Watch it here:
Why Forge: Adds sources beside claims so Forge can flag doubtful facts and send them to Cerebra.
Adds footnote-style sources to research results so you can see where facts came from.
Suits: Cerebra. Cerebra's source-graded research needs citations.
In your videos: Named in 1 of your videos
Watch it here:
Why Forge: Reads your files to learn how the design is built before redesigning them.
Redesign skill that reads your files to learn how your design is put together.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Private wellbeing companion. Stays inside her own walls. 16 picks chosen for this lane. These are suggestions only. Nothing is installed until Sentinel has looked and you say yes.
Why Mantis: Free local voice cloning and transcription, giving her a steady, soothing voice kept on his own machine.
Free local voice cloning, scripts, dubbing and transcription. An alternative to ElevenLabs.
Suits: Mantis, Alfred. Mantis does image prompts and creative work, which this produces; Alfred could also use it.
In your videos: Claimed 61,000 stars · Named in 4 of your videos
Watch it here:
Why Mantis: Free voice dictation on his Mac, so he can speak his thoughts to her easily.
Free voice dictation that runs on your Mac.
In your videos: Named in 1 of your videos
Watch it here:
Why Mantis: Holds her warm voice, boundaries and way of caring in one file, so she stays herself.
A file that tells each agent who it is and how to behave.
Suits: Forge. Forge maintains the agents' own files.
In your videos: Jack Roberts calls it a character bible and gives a template · Named in 10 of your videos
Watch it here:
and 6 more videos
Why Mantis: Lets her remember what helps Chris at the start of each private chat.
Two notes files Hermes reads at the start of every session so it remembers you.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Metics Media had the agent build a NOTES_INDEX.md as well · Named in 9 of your videos
Watch it here:
and 5 more videos
Why Mantis: Lets Chris shift her mood for one chat, gentler or more playful, without changing who she is.
Switches the persona for one session without changing the SOUL.md file.
Suits: Mantis, Alfred. Mantis does image prompts and creative work, which this produces; Alfred could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Mantis: Keeps her keys, memory and persona in a separate private setup apart from the other agents.
Separate Hermes setups, each with its own keys, memory, scheduled jobs, skills and personality.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Alex Finn recommends at least two profiles so one can fix the other · Named in 6 of your videos
Watch it here:
and 2 more videos
Why Mantis: Lets her look back at past private chats to recall what soothed him before.
Searches your old past conversations with the agent so you can find what was said.
Suits: Alfred, Brainiac. Alfred and Brainiac can recall earlier chats on demand.
In your videos: Named in 5 of your videos
Watch it here:
and 1 more videos
Why Mantis: Lets Chris talk with her out loud when typing feels like too much.
Talk to Hermes out loud and use a wake word to hand it tasks.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
Watch out: Alex Finn says ChatGPT voice mode needs an OpenAI API key, about 5 cents a minute
In your videos: Named in 3 of your videos
Watch it here:
Why Mantis: Gives her own personality, memory and model, so she stays distinct and fully private.
Runs several specialist bots, each with its own personality, skills, memory and model, that can mention each other.
Suits: Forge. Forge runs the agent team; bot mode is how the seven agents could share rooms.
In your videos: Sharbel calls it a cheat code; Wanderloots suggests keeping 3 to 5 bots warm and a 10 to 30 minute idle timeout · Named in 3 of your videos
Watch it here:
Why Mantis: Helps capture how Chris thinks and talks, so her replies feel natural and kind to him.
A doc built by interview about how you think, plus a voice guide per medium with banned words.
Suits: Alfred, Mantis. Alfred, the general assistant, could use it in everyday chats; Mantis could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Mantis: A simple map of memory, audio and guardrails that helps shape her calm, safe design.
The six parts of an agent: model, tools, memory, audio, guardrails and orchestration.
In your videos: Named in 1 of your videos
Watch it here:
Why Mantis: A small voice that runs on its own machine, so she can speak calmly without sending words out.
A tiny, fast text-to-speech voice that runs on your own machine.
Suits: Alfred. Alfred could voice replies cheaply, related to Chris's voice plans.
In your videos: Komputer Mechanic found it about 2x faster than Kokoro · Named in 2 of your videos
Watch it here:
Why Mantis: A local voice with a softer sound, so late-night talks can be heard, not only read.
Local text-to-speech voice. Slower than Piper.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 1 of your videos
Watch it here:
Why Mantis: Turns Chris's spoken words into text on-device, so he can vent by talking when tired.
An open model that turns speech into text, so you can talk to your agents.
Suits: Alfred. Alfred, the general assistant, could use it in everyday chats.
In your videos: Named in 2 of your videos
Watch it here:
Why Mantis: A free voice model for many languages, offering her another gentle spoken voice option.
Free voice model covering 600+ languages, used behind Voice Studio.
Suits: Mantis. Mantis does image prompts and creative work, which this produces.
In your videos: Named in 1 of your videos
Watch it here:
Why Mantis: Smooths stiff, machine-like wording so her replies sound warm, short and like a real person.
Rewrites text to remove the usual signs that an AI wrote it.
Suits: Alfred, Mantis. Alfred and Mantis write text for Chris; it is already built into Hermes.
Watch out: Teku AI says it is token heavy; Sharbel notes the skill file is 29.6 KB
In your videos: Playlist-1 doc notes it is already built into Hermes · Named in 4 of your videos
Popularity: 60.1k stars
Watch it here:
Security reviews. Looks and reports, never installs. 33 picks chosen for this lane. These are suggestions only. Nothing is installed until Sentinel has looked and you say yes.
Why Sentinel: Free monitoring of Linux, Docker and websites; a model for watching agent health and watchdogs.
Free monitoring for Linux, Docker, Proxmox and websites, with checks for agent health.
Suits: Cyborg, Sentinel. Cyborg runs the LAN telemetry dashboard and Sentinel watches health; fits Chris's planned agent dashboard.
In your videos: NetworkChuck uses it to monitor his Hermes agents (crons, gateway, model login) · Named in 1 of your videos
Watch it here:
Why Sentinel: Includes a security audit command, worth reviewing as a reference for audit checks.
Local agent platform you install with one command. It includes a security audit command.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Alex Finn suggests it as a backup agent; Greg Isenberg's guest switched away from it · Named in 18 of your videos
Popularity: 393.0k stars
Watch it here:
and 14 more videos
Why Sentinel: A central place for secrets; Sentinel can compare it against keys found in plain sight.
One safe place to store your API tokens. It syncs to Vercel and has a skill.
Suits: Sentinel. Sentinel would keep secrets out of chats and Discord.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: A private encrypted network supporting a zero-trust perimeter; Sentinel can review its access rules.
A private, encrypted network so your agents can reach and log into all your devices.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 8 of your videos
Popularity: 36.3k stars
Watch it here:
and 4 more videos
Why Sentinel: Scans proposed agent skills for security problems, giving Sentinel a repeatable check before installation.
Scans agent skills for security problems before you install them.
Suits: Sentinel. Sentinel vets skills, so a scanner is a direct fit.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: Tests MCP servers so Sentinel can see what a proposed server actually exposes.
Tests MCP servers. Run it with npx @modelcontextprotocol/inspector.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Popularity: 10.7k stars
Watch it here:
Why Sentinel: Flags unused MCP servers, which Sentinel can report as needless exposure.
Shows where your Claude tokens go and flags MCP servers you never use.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: A review point: confirm agents must ask before spending, sending or deleting anything.
Makes the agent ask you first before it spends money, sends things or deletes anything.
Suits: Sentinel, Forge. Sentinel enforces safety and Forge runs ops; gates match Chris's no-surprise rules.
In your videos: Named in 2 of your videos
Watch it here:
Why Sentinel: A health-check command that reports whether each Hermes install is working properly.
A health-check command that reports whether your Hermes install is working.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Why Sentinel: Lets Sentinel see sessions, keys, channels and logs in one place for read-only review.
A web page to see and manage your Hermes sessions, models, jobs, skills, keys, channels and logs.
Suits: Forge, Cyborg. Forge builds and runs agent operations, where this belongs; Cyborg could also use it.
In your videos: Named in 6 of your videos
Watch it here:
and 2 more videos
Why Sentinel: Keeps agent code walled off in Docker, a boundary Sentinel can check is in place.
Runs agent code and browsing inside Docker so it stays walled off from the rest of your server.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Why Sentinel: Separate keys, memory and jobs per agent; Sentinel can verify the separation holds.
Separate Hermes setups, each with its own keys, memory, scheduled jobs, skills and personality.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Alex Finn recommends at least two profiles so one can fix the other · Named in 6 of your videos
Watch it here:
and 2 more videos
Why Sentinel: Turning off unused tools shrinks the attack surface, so Sentinel can check what is enabled.
Turns off tools and built-in skills you do not use, to cut cost and wrong tool loops.
Suits: Forge, Sentinel. Forge builds and runs agent operations, where this belongs; Sentinel could also use it.
In your videos: Named in 2 of your videos
Watch it here:
Why Sentinel: Extra credentials mean more keys to track, so a good item for key audits.
Backup models and accounts so your agent keeps working when one provider fails.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 2 of your videos
Watch it here:
Why Sentinel: Script-only scheduled jobs make simple watchdog checks cheap and predictable to review.
A scheduled-job option that runs only a script, saving tokens by skipping the AI.
Suits: Brainiac. Brainiac's scheduled vault jobs could run as plain scripts.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: Gives Sentinel a clear standard when hunting for tokens pasted into Discord, notes or logs.
Never paste keys into chat. Set them as environment variables in the terminal and give least access.
Suits: Sentinel. Sentinel enforces Chris's no-secrets rule.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: A least-access standard to measure keys and database users against during audits.
Give agents read-only keys and database users that cannot see tables holding personal data.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: A checklist of sudo user, SSH keys and firewall rules to compare the VPS against.
Basic server lockdown: a sudo user, ed25519 SSH keys, and a firewall allowing only SSH, 80 and 443.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: A model for recurring read-only security checks, to propose to Chief for approval.
A scheduled job that checks your agent's security every hour or twice a day.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 3 of your videos
Watch it here:
Why Sentinel: A test for whether any dashboard is openly reachable from the internet.
Open the Hermes dashboard only through a VPN or SSH tunnel, never openly on the internet.
Suits: Sentinel, Cyborg. Sentinel reviews and security-vets skills, which this supports; Cyborg could also use it.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: A rule to check, so only Chief can run admin commands on bots.
When a Telegram bot is open to everyone, allow admin commands for one Telegram ID only.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: Shows agents only allowed API routes, a pattern Sentinel can recommend for tighter access.
A self-built MCP that shows agents only the API routes you allow, such as Redmine.
Suits: Sentinel, Forge. Sentinel would want least-privilege API access for every agent.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: Daily GitHub backups of the vault; Sentinel can confirm backups are actually running.
Saves versions of your agent or vault, with .gitignore, and backs it up daily to GitHub.
Suits: Cyborg. Cyborg already publishes the vault to GitHub.
In your videos: Named in 3 of your videos
Watch it here:
Why Sentinel: Agents logging their actions gives Sentinel a trail to read when checking behaviour.
Each agent writes down what it did, or adds it to today's daily note, before replying.
Suits: Brainiac. Brainiac files and maintains the vault and runs cron jobs, which this supports.
In your videos: Named in 3 of your videos
Watch it here:
Why Sentinel: A paste-in audit of context, memory files and schedules, useful as a review checklist.
A paste-in prompt that checks Claude Code context, deferral, memory files, cache and schedules.
Suits: Forge. Forge builds and runs agent operations, where this belongs.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: Instruction files agents obey; Sentinel can check them for risky or injected text.
A file of instructions Claude reads. Keep it a short map to other files, and check it now and then.
Suits: Forge, Brainiac. Forge builds and runs agent operations, where this belongs; Brainiac could also use it.
Watch out: Mason Fitzgerald calls it a soft instruction that is not enforced
In your videos: Named in 7 of your videos
Watch it here:
and 3 more videos
Why Sentinel: Multi-agent review of code for security problems, helpful when vetting repos Cyborg proposes.
Anthropic's multi-agent review that checks your code for security problems.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 2 of your videos
Watch it here:
Why Sentinel: Shows health checks across several agents, helping confirm every agent is covered by watchdogs.
A dashboard for several Hermes agents showing tasks, health checks and live costs.
Suits: Cyborg, Forge. Cyborg runs the dashboard side and Forge runs fleet ops; matches Chris's planned agent dashboard.
In your videos: Sharbel presents it as his own tool for running a Hermes fleet · Named in 2 of your videos
Watch it here:
Why Sentinel: Sealed-off workspaces for agents, a safe way to test repos without touching real machines.
Safe, sealed-off workspaces from Dagger where an agent can work without touching your real machine.
Suits: Sentinel, Forge. Sentinel reviews and security-vets skills, which this supports; Forge could also use it.
In your videos: Named in 1 of your videos
Popularity: 4.0k stars
Watch it here:
Why Sentinel: Checks a new skill for hidden bad instructions before Chief installs it, fitting her vetting work.
Checks a new skill for hidden bad instructions before you install it.
Suits: Sentinel. Sentinel already security-vets skills; Bouncer automates that first check.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: Turns an agent into a security auditor, useful for structured reviews and written findings.
Turns a coding agent into a security auditor.
Suits: Sentinel. Sentinel reviews and security-vets skills, which this supports.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: Security review skills that can run on a schedule, matching her routine audit checks.
Security review skills, said to be by Google, that can run on a schedule.
Suits: Sentinel. Sentinel does security review; note the name clashes with Chris's Mantis agent.
In your videos: Named in 1 of your videos
Watch it here:
Why Sentinel: Limits which vaults an agent can touch, a control to verify against the vault rules.
Script and skill that limit which vaults an agent is allowed to touch.
Suits: Sentinel, Brainiac. Sentinel and Brainiac could enforce the archive-folder ban with an allow-list.
In your videos: Named in 1 of your videos
Watch it here:
Add-ons that change how Claude Code, Hermes, ChatGPT and Obsidian look or behave: mods, hooks, plugins and themes, grouped by the tool they change.
From your videos: Fed by 25 of the 105 videos; one video (9 NEW Claude Mods) supplies most Claude Code entries. No video covered status lines, output styles or Hermes themes, so 10 entries from the web (Anthropic and Hermes docs, plus community status lines) fill those gaps (marked From the web). Updated 26 Oct 2026: 2 entries from full transcripts of IbFaY3xFpZM and x2F3fdn82GU.
Plugin connectors for Gmail, Notion, Slack, Drive, Calendar, Supabase and Netlify inside ChatGPT or Codex. Riley Brown also shows a Messages integration that reads the local iMessage database.
Watch out: Reading iMessages gives the agent your private chats.
A Chrome extension and an 'any app' setting that let ChatGPT or Codex control the browser and the computer. Jack Roberts calls ChatGPT the current champion at computer control.
Watch out: It drives your real computer; keep it away from accounts that hold money.
A ChatGPT settings toggle called 'show pet'. Dan Martell lists it next to the 'any app' computer-use toggle.
Watch out: Named only; the video does not explain what it does.
Grok Bot opened a marketplace where products become connectors the bot can find and add when it lacks a skill. The Next New Thing's guest is submitting a plugin that runs his AI-assessment workflow.
Pika turns product photos into videos inside ChatGPT; a meetings plugin writes notes and summaries.
Anthropic plugins for finance, HR, legal, marketing and operations work, including legal skills.
For plugins in Anthropic's official marketplace, the details pane shows a Context cost: tokens added every turn and when invoked. After install, the Installed tab groups plugins you have not used recently, so you can switch them off.
For you: Forge's cost watching: drop plugins that cost tokens and are never used.
A marketplace of skills, agents, MCPs and plugins for Claude Code.
Watch out: Community content: review each item before installing.
Anthropic's official Claude Code plugin repository, named in The Next New Thing's roundup.
For you: The safest place to look first, since the author is known.
Hooks are set in settings files (user, project or local), in plugins, or in a skill's frontmatter. Events include PreToolUse, UserPromptSubmit, Stop and SessionStart. A hook that exits with code 2 blocks the action and sends its message back to Claude; other error codes let the action go ahead.
For you: The right place for Chris's never-do rules, such as archive-not-delete.
Watch out: Hooks run shell commands with your own permissions. A failing hook does not block by default.
Hooks are small scripts that fire on events, for example when Claude says it is finished or before it sees your prompt. Simon Scrapes calls them deterministic checks; Sharbel A. uses an output-filter hook that trims command output before the agent reads it.
For you: Put must-never rules in hooks, not in CLAUDE.md, which Mason Fitzgerald calls a soft instruction.
A marketplace is a catalogue of plugins. Anthropic's official one (claude-plugins-official) is added for you the first time you start Claude Code in a terminal. Others you add yourself; their plugins then appear in the /plugin panel's Discover tab.
For you: Sentinel reviews any non-official marketplace before it is added.
Watch out: Plugins can run hooks and MCP servers. Auto-update is on for official marketplaces and off for community and third-party ones.
Each Artcraft app ships with its own MCP (a plug that lets Claude call the app's actions directly). You paste the app's MCP install line into Claude. Then Claude sends exact actions like 'set this layer to 60% opacity' instead of looking at the screen and clicking, which is faster and has fewer misclicks.
For you: Forge, if Chris tests Artcraft.
Watch out: Early software; check every import and result by hand.
A hook on UserPromptSubmit that looks up the 8 nearest chunks in a vector database and adds them before Claude sees the prompt. Mason Fitzgerald claims 5.2 times faster and 3.5 times cheaper than file search on 15 questions.
For you: Relevant to the Qdrant vault-memory plan.
Watch out: His package is behind a paid community and works in Claude Code only. Test on Chris's own vault before believing the numbers.
Runs the same job with and without a skill to show whether the skill actually helps. Simon Scrapes suggests writing skills only for what Claude gets wrong.
For you: Sentinel and Forge: keep a skill only if it earns its tokens.
A marketplace is just a folder or Git repo with a .claude-plugin/marketplace.json file listing your plugins. It can be private, and anyone with access adds it with one command.
For you: A clean way to keep Chris's own reviewed mods and skills in one private GitHub repo.
Built by Anthropic. Before a risky command such as deleting a folder, it stops, lists exactly which files would go, and waits for you to press Proceed or Cancel.
For you: Matches Chris's archive-not-delete rule; the idea suits Sentinel's approval gates.
Warns you before a send that would cost a lot because the prompt cache has gone cold (about an hour on a normal plan), with an estimate in tokens and credits. Typing /keepwarm pings the session just before the cache goes cold.
For you: Forge (cost watching).
Open-source theme mod with seven themes (such as Tokyo) that re-skins tool calls with icons and colours. Jay E calls it a safe first mod because it only changes the display.
For you: Colour-coded output suits Chris's preference for colour.
Adds a file panel on the right. A file shimmers while Claude touches it and turns green once the change is committed. Costs no tokens.
When you correct Claude with words like 'no' or 'this doesn't look good', a small band asks whether to save that as a permanent rule in CLAUDE.md or a skill. One click saves it.
For you: Saves Chris repeating his working rules.
Anthropic's own sample mod. After each turn a replay button steps through every edit Claude made, for code or for a written draft.
For you: Helps Sentinel review what changed.
Puts a progress bar above the prompt and, for big jobs, a live panel of every sub-agent with its model, progress and rough cost. You can restyle the little pixel mascots.
For you: Same idea as Chris's planned agent dashboard: who is working and what it costs.
Opens a real web browser beside the chat in the terminal, so you can check a page or a GitHub pull request without leaving.
A mod is a small add-on that changes how Claude Code looks and works: extra buttons, side panels, even a mini-game while Claude works. A skill tells Claude how to do a job and a connector plugs in another app; a mod changes Claude Code itself. It listens for events (such as Claude running a command) and can watch, change or handle them. Jay E says mods arrived on 1 October and work in the terminal and the desktop app's code tab.
Watch out: Mods are temporary until saved as a plugin. Not in the VS Code extension when recorded. Any mod that calls a model uses your plan's usage.
Built by Anthropic. A second agent reads Claude's replies and pops up anything important you might miss in a wall of text, such as a cost. You can tell it what to watch for.
Watch out: It runs a second agent, so it uses extra tokens.
Ready-made status lines named in Anthropic's own docs. ccstatusline offers widgets (git, usage, a Claude service-health widget) and Powerline colours; starship-claude uses the Starship prompt tool for the status line.
For you: Quicker than writing a script; Sentinel reviews first because it runs on every update.
Watch out: Pick one status line; they replace each other.
A vim-style Powerline status line with live usage tracking, git info and custom themes. A setup wizard walks you through theme, style, font, segments and budget.
Watch out: Needs Node.js 18+ and Git; for the best look install a Nerd Font, or use the plain-text option.
An output style sets Claude's tone and response format for a whole session. Besides Default there are four built-ins: Proactive (starts work without routine questions), Concise (result first, no recap), Explanatory (adds short Insight blocks) and Learning (leaves small pieces of code for you).
For you: Concise matches Chris's short-replies rule; a custom style could hold his neurodivergent-friendly format.
Watch out: A style is an instruction, not a guarantee. Anything that must always happen belongs in a hook.
A custom bar at the bottom of Claude Code that runs your own script. Claude Code sends the script session data (model, context use, cost, git and more) and shows whatever it prints. It runs locally and uses no tokens.
For you: Colour-coded context and cost at a glance, which suits Chris's colour preference.
Watch out: It runs a shell command, so it stays blank until you accept the folder's trust prompt.
Bouncer checks a new skill for prompt injection before install and then allows, asks or quarantines it. Context Doctor audits bloat in the soul file, memory, skill descriptions, tools and config, and quarantines rather than deletes.
For you: Bouncer fits Sentinel's review-before-install rule; Context Doctor fits the archive-not-delete rule.
Watch out: Sentinel should still review both.
Alex Finn turns on Hermes plugins for Browser Use, Browserbase, Firecrawl and OpenAI image generation, and toggles browser automation in 'hermes tools'.
For you: Cerebra (web research).
A bundled Hermes skill that helps you write your own colour theme as one YAML file in ~/.hermes/skins/. The same file themes the CLI, the TUI and the desktop app.
For you: Could give each agent Chris's own colour code.
Watch out: The skill suggests at least about 4.5 to 1 contrast between accent and background, for readability.
Komputer Mechanic built a web dashboard over Hermes with mission, documents, schedule and terminal tabs and an 'Agent City' view of who is busy. Julian Goldie's Agent OS watches Claude, Codex and Hermes agents with steer and stop buttons.
For you: Ideas for Chris's own agent dashboard.
Watch out: Both are paid: Komputer Mechanic's 'free prompts' need a $10-a-month membership; Agent OS is sold in a paid community.
Lets Hermes control DaVinci Resolve Studio live.
For you: Forge (video editing).
In the Hermes desktop app, Capabilities > Plugins > Browse lists official and community plugins, with badges for reviewed ones. Teku AI shows it in the big desktop update.
For you: Start here before any GitHub plugin; Sentinel still reviews.
Skins change only how Hermes looks: colours, spinner, agent name, welcome text, prompt symbol, tool emojis and banner art. Built-in skins include default, ares, mono, slate, daylight, warm-lightmode, poseidon, sisyphus and charizard.
For you: A different skin per agent would make the seven agents easy to tell apart.
Watch out: Skins do not change personality or tone; that is a separate setting.
Rewrites an agent's draft so it sounds less like AI. Dubibubi says it is built from Wikipedia's 'Signs of AI writing' page and updates when that page changes. It is already built into Hermes, but few people switch it on.
For you: Alfred's briefings or any public text; use the built-in copy, don't install a second one.
Watch out: Other videos call it token heavy (the skill file is about 29.6 KB). Dubibubi's own tip: a good skill stays under 15 KB.
/kanban create builds a task board with researcher, writer and reviewer bots. Metics Media also shows Achievements.
For you: The warnings page says try this board before adding Paperclip.
Animated avatars, such as Pikachu, for Hermes bots in bot mode. Wanderloots shows them.
For you: A fun visual touch for the seven agents; no cost noted.
A community plugin that lets Hermes run approved scripts on a real Android phone or a mirrored iPhone; install with 'hermes plugins install phone-harness'.
Watch out: It is on by default in messaging apps, so anyone who can message the bot could control the phone. Needs Hermes 0.21.6 or newer. The warnings page says switch it off per platform or skip it.
A plugin that sends only the relevant skills to the model. AI Labs shows skill context drop from about 11,000 to about 2,300 tokens.
For you: Forge: a direct saving on the Haiku budget.
Watch out: No owner/repo stated.
A library of community Hermes skills. Metics Media says to review anything from it before use.
Watch out: Untrusted until Sentinel has read it.
A community plugin for chatting with AI over your vault, named alongside Smart Connections.
For you: Brainiac.
Watch out: Pick one of the two; both do a similar job.
A custom mini-app inside Obsidian with focus, metrics, a heat map and a quick-capture inbox.
Watch out: Sold through his community, not free.
Clips web pages and YouTube transcripts into the vault's raw folder, ready for an agent to file. Wanderloots uses it to feed his LLM wiki.
For you: Feeds Brainiac's filing with less copy and paste.
A community Obsidian plugin that connects AI to your vault. Tina Huang names it as one way to bring AI into Obsidian.
For you: Brainiac and Cerebra, inside The Cone of Silence.
What the videos mean by an "AI OS" or "Agent OS": one layer or dashboard where your agents, memory, tasks, schedules and tools run together. This tab shows each version the presenters built, the parts they share, how to build one, and what fits the Gauntlet.
From your videos: 44 of the 105 videos feed this tab. Only a few use the words "operating system" (Pav Rusovs, Jack Roberts, Julian Goldie, Max Mitcham, Tina Huang's HermesOS); the rest show the same idea as a mission control, dashboard, Kanban board or vault. Most finished AI OS builds shown are paid or unreleased, so the useful part is the pattern, not a product to install. Plus 14 entries from the web to fill gaps (marked From the web): Hermes's own dashboard, Kanban board and tutorial, Langfuse and OpenTelemetry tracing, Open WebUI as a front end, four open-source dashboards that exist today on GitHub, and four outside descriptions of an AI OS. The Paperclip entry now links its GitHub repo (its README lists Hermes as supported).
The tutorial's example uses two profiles, one doing work and one checking it, with the second task waiting on the first.
For you: A test pair like Cerebra (research) then Brainiac (file to vault) would show the flow.
Watch out: The tutorial does not explain how to create the profiles themselves.
One board shared by all Hermes profiles, stored in ~/.hermes/kanban.db. Columns are triage, todo, ready, running, blocked, review, done and archived. A dispatcher inside the gateway checks every 60 seconds, moves todo to ready when parent tasks are done, and starts the assigned profile as its own process. Workers finish with complete, request review or block.
For you: Matches the donts.json advice: try this native board before Paperclip or any meta layer.
Watch out: Designed for one trusted host; avoid dependency loops, and note scratch workspaces are deleted on completion unless outputs are declared as artifacts. With hermes dashboard --host 0.0.0.0, every plugin route, the board included, is reachable from the network.
Checked 26 Oct 2026
A community OpenTelemetry plugin (Apache-2.0) that exports each turn's model calls, tokens, cost, tool calls, sub-agents and approval waits. Tested with Phoenix, Langfuse, SigNoz, Jaeger, Grafana and others. With no backend set it keeps a local SQLite store and sends nothing.
Watch out: Community tier, and content_capture defaults to full prompts and replies; set preview or off. Sentinel review first.
hermes dashboard opens a local web app at 127.0.0.1:9119. Its pages: Status, Chat, Config, API Keys, Sessions, Logs, Analytics (tokens and cost over 7, 30 or 90 days), Cron, Profiles, Skills, MCP, Webhooks, Pairing, Channels and System.
For you: Analytics already shows cost per Hermes home; compare it with what the Gauntlet plans to show before building a duplicate.
Watch out: Login is off on loopback and forced on for any other address; keep it on Tailscale or an SSH tunnel.
Checked 26 Oct 2026
An official Hermes plugin sends one trace per turn to Langfuse, with each model call's tokens and cost, each tool call and sub-agent, and tags for platform, profile, cron job id and Kanban task. Langfuse itself is open source (MIT except its ee folders, about 36.5k stars) and can be self-hosted with Docker Compose.
For you: Per-profile and per-cron cost tags are what the Gauntlet's cost page needs; metadata mode suits Mantis.
Watch out: Default mode is sanitized, which the docs call not a DLP guarantee; metadata mode sends no content. Telemetry can be lost if the queue fills.
Hermes has an OpenAI-compatible API server. Open WebUI connects to it like any OpenAI endpoint, Hermes runs its tools, and the answer streams back. The model appears as hermes-agent in Open WebUI's list.
For you: Chris already runs Open WebUI on Coolify, so this gives one chat screen for the agents.
Watch out: In Docker on Linux, localhost and host.docker.internal may not reach Hermes; the docs list fixes.
An open-source research project that builds a kernel for agents: it schedules model calls and manages context, memory, storage and tools, with a separate Cerebrum SDK. Paper accepted to COLM 2025. About 6.5k stars; some deployment modes are still in development.
Watch out: A research codebase, not a ready dashboard for a Hermes fleet.
Checked 26 Oct 2026
In his talk "Intro to Large Language Models", Andrej Karpathy says LLMs can be thought of as the kernel process (the core program) of a new kind of operating system. The context window is like RAM, and the model pages information in and out of it, while using tools like a browser, files and other software.
A 2023 paper by Ge and others maps an operating system onto AI: the LLM is the kernel, the context window is memory, outside storage is the file system, tools are devices and libraries, prompts are user commands, and agents are the apps.
For you: In Chris's terms: OpenRouter models are the kernel, the vault is the file system, and the seven agents are the apps.
The MemGPT paper (Packer and others, UC Berkeley) borrows virtual memory from operating systems: it moves information between fast and slow memory tiers so the model seems to have more context than it does, and uses interrupts to manage control flow.
For you: The same tiering idea as Hermes memory plus the vault; useful background, not something to install.
Checked 26 Oct 2026
This order is pieced together from the MAPS layers, the Hermes level videos and the mission-control builds. Each step works on its own before the next is added.
The presenters add the same guard rails. Pav Rusovs blocks send, share and delete on the server so the agent can only draft email. Jack Roberts never pastes keys into chat and does not let Hermes send email. Nimrod gives agents scoped, read-only keys. Wanderloots caps turns in group rooms.
For you: Matches Sentinel's review role and Chris's no-secrets-in-Discord rule.
Not an app but a GitHub template: a documentation folder for a Hermes fleet on a VPS, with a system map, agent registry, runbooks, secret map (names only, never values) and recovery notes. It describes four growth levels, from one agent to an automated team, and a task bus pattern. MIT, about 732 stars.
For you: The same idea as Chris's vault infrastructure notes; worth comparing its folder list, not installing.
Watch out: Its setup skills install Node.js, Claude Code, Codex, Docker and Hermes on a server; Sentinel must read them first.
Checked 26 Oct 2026
A browser interface for Hermes with sessions, chat, a workspace file browser with Git status, Cron/Tasks, Skills and Memory panels, profiles and voice input. Aims for near parity with the Hermes CLI. MIT, about 19.2k stars.
For you: Likely the Hermes web UI Chris already runs on Coolify.
Watch out: Password is off by default; set HERMES_WEBUI_PASSWORD or a passkey.
Checked 26 Oct 2026
A self-hosted control plane for agents: tasks with review and a quality gate, agent registry, schedules, alerts, logs, token use and cost views, memory browser, skills registry, roles and approvals. MIT, about 6.4k stars. Supports OpenClaw, Claude Code, Codex, CrewAI, LangGraph, AutoGen and Claude SDK.
Watch out: Hermes is not on its supported list, and the README calls it alpha software whose APIs may change.
Checked 26 Oct 2026
A Kanban board that runs coding agents (Claude Code, Codex, Gemini CLI, GitHub Copilot, Cursor and others) in their own branch and terminal, with diff review and pull requests. Apache-2.0, about 28.5k stars.
Watch out: The README opens with "Vibe Kanban is sunsetting"; do not start on it.
Checked 26 Oct 2026
Each agent gets one job. Dan Martell's AGENT framework (Aim, Give identity, Equip, Narrow, Trust) gives each agent a lane and a one-sentence definition of done, with a manager agent delegating. Alex Finn keeps at least two profiles so one can fix the other.
Sharbel A.'s dashboard shows live costs; Jack Roberts sees spend across every model in one view. Jack Roberts and Metics Media both use per-key spend limits on OpenRouter.
Every version starts with plain markdown: SOUL.md for personality, USER.md and MEMORY.md for facts. Tina Huang lists four Hermes memory tiers: markdown files, session search, an external provider such as Honcho, then an Obsidian vault. Sharbel A. says only one external provider can be active at a time.
Watch out: Do not run two external memory providers at once.
NetworkChuck monitors his Hermes agents in Checkmk: crons running, gateway up, model login hours left, whether she is replying, and usage, through a small local script on each agent's VM. When the gateway stops, an alert goes to Slack and recovery is reported when it comes back. Tina Huang uses an alerts channel fed by cron jobs.
For you: A gateway-up and model-login check per agent is a natural Fleet tile on the Gauntlet.
The OS runs on a heartbeat: cron jobs, routines and morning briefs. Sharbel A. says each cron job should be self-contained and return a decision. Alex Finn says to prune and pause old jobs. Cron memory lets jobs remember past runs.
The dashboard shows status, tasks, documents and costs. Pav Rusovs keeps it read-only. Komputer Mechanic adds a Documents tab so long reports leave the chat and save tokens.
A Kanban board shows who owns what and what needs approval. Hermes has one built in; Paperclip, Buzz and Multica offer bigger ones.
A dashboard over Claude, Codex, Hermes and Antigravity, with a "memory galaxy", automations listed inside it, and a view to watch sub-agents with steer and stop buttons. He runs the Hermes phone-control plugin inside it.
Watch out: Sold inside a paid community. The phone-control plugin shown with it is on by default in messaging apps, so others messaging the bot could drive the phone.
Sharbel A. calls Hermes Desktop the control room. It has sessions, pinned chats, profiles, an artifacts tab for every file sent, a skills panel to switch off unused skills, a cron panel, and an agents window. Newer versions add sub-agent supervision (view, stop, steer), split sessions with one composer, and a HUD that sees your screen.
His level-seven "agentic operating system" shows connected brands and integrations, a Hermes chat, mid-term goals broken into steps for Hermes and for him, the Pantheon skill roster (each skill has a job, description, optional system prompt and its own model), a memory section joining Obsidian, Claude Code and Hermes memory, a documents area, and spend across every model. "Hermes dream" reviews his activity across tools and suggests improvements.
For you: The per-skill model choice and the spend-across-models view map to the Gauntlet's Costs and Keys pages.
Watch out: The OS itself comes with his paid course and community; copy the layout idea, not the product.
Her setup runs several Hermes agents with clear lanes: Coder (takes a spec in Discord, Claude Code builds, a local Qwen model reports back), Lifebot (health and productivity coach on Telegram), Taco Bot (operations assistant) and an alerts channel fed by cron jobs. The native Hermes Kanban board orchestrates work, and her own apps log into an Obsidian vault.
For you: Closest to Chris's own setup: one channel per agent in Discord, a vault for logs, and cron-driven check-ins.
Hermes already ships a basic AI OS. The `hermes dashboard` command opens a web UI with sessions, models, cron, skills, keys, channels, logs, plugins and profiles. The Kanban board (triage, to do, ready, in progress, blocked, done) splits work across sub-agents and profiles, and Wanderloots uses it for ownership, revisions and approvals.
For you: Try the native board before adding Paperclip or any other meta layer.
Watch out: Lead Gen Jay says he never logs into the dashboard and drives Hermes from Claude Code instead; the dashboard is optional.
An old PC runs Claude Code, Codex and OpenCode around the clock, reached over SSH and Tailscale. T3 Code is the control surface on laptop and phone. An "HQ" project, the same repo on every machine, lists all projects and starts threads remotely; a shared wiki is the memory; Hermes takes chat and voice notes from an iPhone Action button.
For you: His scoped, read-only keys per agent match Sentinel's least-access approach.
Four layers: Memory (a short CLAUDE.md map of about 40 lines plus signpost index files), Agent (Claude Code on laptop and VPS, synced every five minutes), Pulse (a routines.yaml list checked by a scheduler every five minutes, with a run cap) and Screen (a read-only live dashboard, "show, don't store"). He measured the map version at 30 seconds and 17k tokens against 62 seconds and 28k tokens without it.
For you: The Screen layer matches the Gauntlet's job: read what n8n and the agents write, never become the place data lives.
Watch out: He chose Claude Code over Hermes because, he says, Anthropic's rules bar using a Claude subscription through third-party harnesses; Chris uses OpenRouter keys instead, so this does not apply the same way.
A web dashboard for 16 Hermes agents. A boss agent plans missions and picks agents; missions move across a Kanban (Briefing, Active, Review, Delivered) with approve, retry and replay. A Documents tab holds long reports so they stay out of chat, a Schedule tab runs Hermes cron jobs, an activity log records each action, "Agent City" shows working or idle agents, and a terminal tab runs Hermes commands without SSH. Voice uses local Piper speech.
For you: The Documents tab and activity log are the two ideas most worth borrowing for the Gauntlet.
Watch out: Despite "FREE" in the title, the dashboard, prompts and installer are paid ($10 a month).
More often Obsidian is one layer, not the OS. Wanderloots builds a three-layer LLM wiki (raw sources, wiki, schema) that an agent ingests, lints and saves to Git. Nimrod points every agent at one markdown wiki with two lines in CLAUDE.md and AGENTS.md: look in the wiki first, write notes to the wiki inbox. Tina Huang ranks Obsidian as the fourth memory tier for Hermes.
For you: Brainiac's filing job already makes the Cone of Silence this layer.
Watch out: Mason Fitzgerald argues agent search over Obsidian is slower and costlier than RAG; test on your own vault before changing.
He runs his whole day in one vault: a dashboard plugin with focus, metrics, a heat map and quick-capture inbox, a /today command that pulls calendar and carry-overs and sets three priorities, a voice "close day" log, an activity log, and terminal, chat and email windows docked inside Obsidian. His backend video treats daily notes as database rows, with a human side agents may not edit and a machine side they can.
For you: The human/machine split matches Chris's rule that agents never read the archive folder.
Watch out: His dashboard plugin and front-end app are sold through his community.
Several presenters build their own: Sharbel A.'s fleet dashboard shows dispatch, health and live costs, and his Hermie HQ is a Google-gated page on Vercel. Tina Huang's mission control has an agent office, tasks, content, calendar, memory, docs and team. Alex Finn has Hermes build one with a content pipeline, memory wiki and docs.
For you: Live costs and health per agent are already on the Gauntlet's plan; these confirm that is what others put first.
Paperclip runs agents as employees on an org chart with tasks, routines (such as daily stand-ups), per-agent secrets, decisions and exportable org settings. NetworkChuck hires remote Hermes agents into it as engineers and later sends monitoring alerts to it as IT tickets that an agent fixes.
Watch out: Add a meta layer like this only if the native Hermes board falls short.
A local app with home cards and themes, workspaces that each get their own brain and memory (with shared core memories), sessions, hire-a-bot teams where each bot has a description and a model, automations that update dashboard cards, built-in docs, sheets and slides editors, and a mobile app to text your agents. Hermes is listed as one supported backend.
Watch out: Not yet released; no repo was given. A cloud version (so it runs with the laptop off) was still in progress.
Samin Yasar calls Buzz an agent OS: a self-hosted workspace with inbox, channels, agents and projects. Kylon and Multica are similar boards where people and agents share channels and tasks. Nate Herk adds the idea of an "agent brief": ask your AI OS to export what it knows about you to onboard a new agent.
Watch out: Kylon's spelling and URL are unconfirmed, and Nate Herk says his video was a sponsored request.
People, agents and a markdown "brain" of context. An hourly routine captures Slack, meetings and email into an inbox, a librarian step files it, and skill chains run work end to end. A Labs page publishes prototypes and test results.
Tina Huang breaks a local agent into seven parts: a home (where it runs), a channel (how you talk to it), a brain (the model), memory, skills, a heartbeat (scheduled runs) and eyes (computer use). An AI OS is these parts made visible and shared across several agents. Her no-code agent video gives a similar list of six parts: model, tools, memory, audio, guardrails and orchestration.
Jack Roberts compares Hermes to a race team and Claude or ChatGPT to the drivers: Hermes is the system around the model. NetworkChuck calls Paperclip a "meta harness" that runs Hermes, Claude Code, Codex and Pi agents as one company. So an AI OS can sit on one harness (Hermes) or above several.
An AI OS is the layer that sits above your agents and ties them together. Jack Roberts describes his top Hermes level as "one operating system for every AI": memory, skills, documents, goals and spend in one place, so tools stop working in silos. Pav Rusovs calls it four layers on top of Claude Code. Tina Huang's version is a team of Hermes agents with shared boards and channels.
For you: Chris already has most of the pieces (seven Hermes agents, the vault, n8n, Discord, the Gauntlet); an AI OS is mostly about joining them up in one view.
The Gauntlet already covers fleet, costs, keys, schedule and alerts. The videos add a few parts it does not have yet: a documents area for long agent reports, an activity log fed by each agent, a health check per agent (gateway up, crons running, model login), and a read-only rule so the page never becomes the store of record.
For you: Fits Forge as builder and Sentinel as reviewer; no new install is needed for the first three steps.
Watch out: Most finished AI OS dashboards shown are paid; build from the pattern rather than buying a product.
Ordered paths from the videos: what to learn first, and what to build in what order, from chatbot to agents to automation. Each road map keeps the presenter's own order, and the last entry is a suggested order for Chris's own setup.
From your videos: 24 of the 105 videos feed this tab, and about 12 give a real ordered path (Tina Huang's skills roadmap, RPG class map, local AI by hardware and ways to run open models; the six- and seven-level Hermes guides; Sharbel A.'s Hermes curriculum; Dan Martell's AGENT framework). Learning paths lean on courses; few videos give a path for someone who already runs a fleet, so the last entry is assembled from several videos. Plus 9 entries from the web to fill gaps (marked From the web): Anthropic's agent guide, Anthropic Academy and its older GitHub courses, two DeepLearning.AI courses, the Hugging Face Agents and MCP courses, roadmap.sh's AI Agents and AI Engineer maps, and an ordered course path for a fleet owner.
Three levels, each building on the last. Basics are for everyone, intermediate is about agents doing your own work, and advanced is for building for other people. She says the basics take days to weeks; real AI coding needs about two to three months of learning to code first.
For you: Chris is already at the intermediate level with Hermes; his CTO background makes the advanced steps realistic.
Watch out: She warns that AI coding on top of weak coding skills does not get far.
Everyone starts on Tutorial Island with one starter course path, then picks one of six classes. She says you only need one course path, not several. Classes: Strategist (AI leadership), Rogue (hacking agents together), Alchemist (reliable agent systems for business), Crafter (AI-built apps), Sage (how models work, fine-tuning) and Machinist (AI on hardware).
Watch out: Vendor courses lean toward that vendor's own products; she notes the Anthropic and OpenAI courses both do this.
For people who set AI direction and lead adoption. Her first pick is OpenAI's free AI leadership course (about 3 to 4 hours), with Andrew Ng's AI for Everyone as an optional add-on that is older but has lasting team frameworks.
For you: The class that matches Chris's CTO role.
For tinkerers who learn by doing. There is no general course; you pick one agent ecosystem (Claude, OpenAI, Microsoft Copilot, or open ones like Hermes and OpenClaw) and work through its docs. Along the way you learn gateways, memory, skills, security and self-hosting models.
For you: Where Chris's seven-agent Hermes fleet already sits.
For turning agents into dependable parts of business workflows. You choose a no-code route (Google Skills Agent Fundamentals or Vanderbilt's agentic AI course) or a code route (deeplearning.ai, Hugging Face agents course, Kaggle's AI agents intensive), then go deep on one platform. Everyone must learn reliability.
Crafters build AI-native apps with tools like Codex, Claude Code, Cursor, Copilot, Lovable or Bolt; she found no thorough general course, so learning is tool-specific. Sages learn how models work (Karpathy's Zero to Hero, a Stanford course, 3Blue1Brown), then fine-tuning or training. Machinists run AI on hardware such as ESP32 and Raspberry Pi.
A beginner mental model for Hermes, taught in this order. His core rule: memory holds facts, skills hold repeatable procedures, and cron jobs should be self-contained and return a decision.
An interview, not a strict course, so the order below follows how the skills build on each other. She names four hiring skills: taste, communication, context engineering and model routing. Her manager keeps one skill file per project and keeps adding context to it.
A ladder you climb without skipping steps, each with a small "level up" task. He hands over trust in a set order: let the agent read and organise for a few weeks, then draft, and sending on its own comes last if ever. When anything misbehaves, check credit, gateway and logs.
Watch out: Switching models mid-conversation makes the new model reread everything, which costs more; each scheduled job also remembers the model it was created with.
From one-shot prompts to an agentic operating system. He estimates savings rise from about 30 minutes a day at level one to about a day a week at level three. Each level adds one habit.
Watch out: He warns against the "MCP candy store": every extra connection grows the context window. He also lets Hermes draft email but not send it.
Five steps to a first agent, with trust widened gradually. Pick work that passes his Rule of R: repetitive, rules-based, and a good return on time.
A research-to-audio agent built in n8n, then tested with evaluations. She frames every agent as six parts and adds them one at a time.
Add memory in this order and stop when it is enough. Sharbel A. adds that only one external provider can be active at a time.
Obsidian basics first, then three levels of letting AI into the vault, ending with a Hermes agent running Karpathy's LLM wiki pattern (raw sources, an AI-kept wiki, a schema file, and ingest, query and lint jobs).
Four main categories plus two advanced bonus ones, presented in this order. She calls the browser playground the easiest of all, and fine-tuning locally one of the hardest jobs.
From microcontrollers to rented data-centre GPUs. Her sizing rule: take 75 percent of RAM as usable, divide by 0.6, and that is roughly the largest model in billions of parameters.
Watch out: Several model names in this video come from auto-captions and may be slightly off; confirm before downloading.
A suggestion, not a plan he has agreed to. Chris is already near the top of the Hermes ladders (seven agents, channels, cron jobs, a vault and a dashboard), so these stages focus on what the videos recommend next for a fleet that already runs. Each stage uses only advice from the cited videos, and nothing gets installed without Sentinel's review and Chris's yes.
For you: Stages 1 to 4 need no new installs; stages 5 and 8 would need Sentinel's review first.
Watch out: This order is assembled from several videos; no single presenter gave a road map for a fleet this size.
Anthropic's December 2024 guide. Workflows follow set code paths; agents decide their own steps. It names five workflow patterns: prompt chaining, routing, parallelization, orchestrator-workers and evaluator-optimizer. Its main advice: use the simplest thing that works and add agents only when needed.
For you: Many of Chris's cron jobs are workflows, not agents; this helps decide which jobs need a model at all.
Watch out: Agents cost more and errors can compound; test in a sandbox with guardrails.
Checked 26 Oct 2026
Anthropic's free course site. Its own course page (checked 26 Oct 2026) lists over 20 free courses, including Claude Code in Action, Claude Code 101, Building with the Claude API, Introduction to Model Context Protocol, MCP Advanced Topics, Introduction to Agent Skills, Introduction to Subagents, and AI Fluency.
Watch out: Titles confirmed on Anthropic's course site. Some courses now say they live on Claude Academy (academy.claude.com).
Checked 26 Oct 2026
Five notebook courses in a suggested order: API fundamentals, prompt engineering tutorial, real world prompting, prompt evaluations, tool use.
Watch out: The repository is archived (read-only) and uses Claude 3 Haiku, so model details are dated.
Checked 26 Oct 2026
An intermediate course of about 10 hours (9 h 55 min, 31 video lessons), taught by Andrew Ng, in plain Python, no framework. Modules: agentic workflows, reflection, tool use, practical tips (evals, error analysis, traces), and patterns for highly autonomous agents.
For you: The evals and error-analysis module is the part a fleet owner gains most from.
Watch out: Videos can be audited free; graded work and the certificate need Pro ($25 USD a month billed yearly, or $30 USD billed monthly).
Checked 26 Oct 2026
A short intermediate course, about 2 hours (1 h 58 min, 11 video lessons), taught by Anthropic's Elie Schoppik. It covers how MCP (a standard plug for tools and data) works, building an MCP server, and connecting it to apps like Claude Desktop.
Watch out: Free during the platform beta; certificates need Pro.
Checked 26 Oct 2026
Free, with free certificates. Units: what agents are, three frameworks (smolagents, LlamaIndex, LangGraph), agentic RAG, and a final project. Three bonus units cover function-calling fine-tuning, agent observability and evaluation, and agents playing Pokemon battles. About 3 to 4 hours a week; needs basic Python.
For you: The observability and evaluation bonus unit is the most relevant part for a running fleet.
Checked 26 Oct 2026
Free, built with Anthropic. Units: MCP basics, an end-to-end MCP app, a deployed use case on Hugging Face, and bonus partner units. About 3 to 4 hours a week; needs some Python or TypeScript.
Checked 26 Oct 2026
Two community roadmaps drawn as clickable diagrams. AI Agents is billed as "Learn to design, build and ship AI agents in 2026". AI Engineer describes someone who uses pre-trained models and existing AI tools to build products.
Watch out: The page text could not be read as a list of stages, so the order of topics is not given here; open the diagrams to see it.
Checked 26 Oct 2026
A suggestion built from the web sources above, not a plan he has agreed to. It skips beginner material (Claude 101, AI Fluency, API basics) because Chris already runs seven agents, and puts the parts that improve a running fleet first: knowing when an agent is needed, skills and sub-agents, MCP, then evaluation and tracing.
For you: Steps 1 to 5 are reading and videos only; step 6 is the first change to the fleet.
Watch out: Most courses use the Claude API directly; Chris's agents use models through OpenRouter, so code examples need that swap.
Complete sets of tools that presenters run end to end, laid out layer by layer: model, harness, memory, tools, hosting and interface.
From your videos: Draws on 44 of the 105 videos; 27 stacks come from the videos that show a whole working setup. Most presenters do not state a monthly total; costs are given only where a video named one. News, repo-roundup and single-tool videos are not stacks and were left out. Plus 5 entries from the web with real published prices (marked From the web): Hostinger Canada prices in CAD, Anthropic and OpenRouter rates in USD, a zero-cost trial stack and a private local stack. Video costs are kept as stated and labelled USD.
This is Chris's setup, not a presenter's. Seven Hermes agents, each in its own lane, on one VPS, with the Obsidian vault as shared memory. It matches the shape many presenters build: a VPS host, OpenRouter, one chat channel per agent, and a markdown vault. Monthly cost: not taken from a video; see the OpenRouter and Hostinger accounts.
For you: This is Chris's own stack. Closest matches in the videos: Tina Huang's HermesOS and Metics Media's beginner Hermes build.
The Obsidian vault is the database and workspace; agents read and write markdown, log to the daily note and stay off the human side. Monthly cost: not stated (his front-end app is sold in a membership).
For you: The human and machine split matches Chris's archive rule; Brainiac's lane.
An open-source Claude Code assistant with voice, a reacting visualiser, a renamable personality and Obsidian-vault memory. Monthly cost: not stated.
For you: A single-user assistant; Chris's fleet already goes further.
Watch out: The exact repo name is not spoken in the video.
Provider-agnostic coding agents on several machines, one shared markdown wiki as memory, and an always-on Hermes reached by voice from his phone. Monthly cost: not stated.
For you: His least-privilege keys are a model for Sentinel's audits.
Claude Code organised into four layers: Memory, Agent, Pulse and Screen. A short map file, a locked-down VPS and a scheduler with a run cap. Monthly cost: not stated (he sells a paid audit service, not the stack).
For you: His map-file test (30 vs 62 seconds) is worth repeating on Chris's vault.
Alex Finn lets an agent act as CEO of a small directory site: it builds the site, runs daily routines, sends a newsletter and takes sponsorship payments, asking approval before external actions. Monthly cost: not stated (Vercel hosting is free).
For you: Not a fit for the fleet; shown for comparison.
Watch out: Dubibubi's 10-day agent shop made 4.55 USD; treat earning claims with care.
Claude Opus 5.5 builds an iOS app, a landing page and a launch video from one project, using design references and a handoff file between chats. Monthly cost: not stated.
For you: Useful for Cyborg's iOS work and Forge's launch videos.
Watch out: Only an Arabic caption track was read; some names are uncertain.
A creator stack built around OpenAI's Codex app, with Hermes added as a separate profile through the Codex login, Obsidian for memory and Remotion for video. Monthly cost: not stated; his Agent OS dashboard is sold in a paid community.
For you: Mostly a showcase; the Remotion and Obsidian parts are the reusable bits for Forge.
Watch out: Captions garbled many names; Agent OS is paid.
A Hermes agent on a cheap VPS, managed from Claude Code with a custom /bob command, tokens in Doppler and apps through Composio. He mentions 35 cron jobs. Monthly cost: Hetzner VPS about 5 to 6 USD a month; Unipile about 5 USD per connected account.
For you: Doppler-style secret storage matches Chris's no-secrets-in-Discord rule.
Watch out: His skill packs are paid.
Eight scheduled cloud workflows in ChatGPT Work using plugins: a commitments tracker, tweet drafts, podcast-guest research and a monthly report. Monthly cost: not stated.
For you: A hosted no-code alternative; no self-hosting.
Watch out: The video is OpenAI-sponsored.
Humans and agents share one self-hosted workspace (Buzz), with each model given a role and subscriptions shared across the team. Monthly cost: not stated.
For you: Interesting if Chris ever adds human teammates; Discord already does the room job.
A content team of three Hermes bots: the main agent as orchestrator, a researcher and a creator. Only the creator is connected to Instagram, and Sharbel A. talks only to the orchestrator. Monthly cost: Hermes is free and open source, so he says the only cost is setup time; models ran on his subscription.
For you: A template for an Alfred-led team; Chris would add Sentinel's review first.
Seven Hermes workflows: meetings to Notion tasks, draft-only inbox replies, a site watcher, a daily client pulse, phone control via a Mac mini and a morning priority brief. Monthly cost: not stated (Mem0 has a free tier).
For you: The draft-only and watcher patterns fit Alfred and Sentinel.
Several Hermes profiles on a mix of paid and local models, a Kanban board fed each morning, a memory wiki site and Tailscale across all devices. Monthly cost: Claude Opus about 40 USD a day (his figure); a DGX Spark for local models costs about 4,800 USD up front.
For you: His two-profile 'doctor' habit and cron pruning both apply to Chris's fleet.
Watch out: His model spend is far above a Haiku budget.
Sixteen Hermes agents run from a custom web dashboard: a boss agent plans missions, agents log actions, long reports go to a documents tab, and there is voice in and out. Monthly cost: the dashboard, prompts and installer sit behind a 10 USD a month membership.
For you: The documents-tab idea fits the Gauntlet dashboard plan.
Watch out: Titled 'free prompts' but the product is paid.
One chief-of-staff Hermes agent managing research, writing and outreach sub-agents, with a cost router sending each task to the cheapest capable model. Monthly cost: not stated (hosting is a sponsor plan).
For you: The cost router is the part to copy for Chris's budget.
A small team of Hermes agents, each with one job: Coder builds software, Lifebot coaches health and productivity, Taco Bot acts as company COO. A separate alerts channel collects check-ins and security reports. Monthly cost: not stated.
For you: The closest match to Chris's fleet; her single alerts channel is a pattern worth copying.
Hermes bot mode with an orchestrator, a researcher and a librarian in group rooms with a turn cap. Research is cited, and the librarian files it into an Obsidian LLM wiki for approval. Monthly cost: not stated.
For you: Maps onto Cerebra (researcher) and Brainiac (librarian).
Watch out: His starter kit and some scripts are members-only.
Not an AI stack: a free home-lab setup for remote video editing, note sync and streaming, all over Tailscale. Included because it shows the private-network layer clearly. Monthly cost: 0 USD, as she frames it.
For you: Shows how Tailscale alone can replace several cloud services.
A Mac mini runs open models offline, with document chat and an n8n-ready interface, reached from an iPad. Monthly cost: not stated.
For you: A route to free local models for simple jobs if Chris buys a Mac mini; size it with Tina Huang's RAM rule first.
Watch out: Better Creating calls Perplexity Computer's hybrid route expensive.
A home lab watched end to end by Checkmk, including a Hermes agent VM. Alerts go to Slack, and one alert type opens a Paperclip ticket that a fixer agent handles. Monthly cost: Checkmk Community Edition is free; other costs not stated.
For you: The Hermes health check (crons, gateway, model login) is a ready template for the Gauntlet.
Watch out: Sponsored by Checkmk; the free edition needs manual plugin work.
Not an AI stack: self-hosted apps that replace big-tech subscriptions for media, books, photos and documents. Monthly cost: not stated.
For you: Coolify already does Portainer's job for Chris.
The cheapest working Hermes build shown: a one-click VPS template, OpenRouter with a weekly spend limit and a cheap tool-using model, reached from Telegram. Monthly cost: Hostinger KVM 1 about 6 USD a month (Hostinger's Canadian page lists KVM 1 at CAN$9.09 a month on promotion, renewing at CAN$16.79) plus model use; Nous Portal is a flat option at about 20 USD a month.
For you: Chris's setup started from the same shape.
A fully offline setup: a local model through Ollama and PrivateGPT for chatting over your own PDFs. Monthly cost: not stated.
For you: A private option; not needed for the current fleet.
Watch out: These videos also promote uncensored models, which remove safety refusals; not recommended.
Tina Huang's free route: open models served locally and an agent built in n8n, with a vector store for documents. Monthly cost: not stated; the parts are free to self-host.
For you: Shares parts with Chris's n8n and Qdrant plans.
A local chat app where chats, users and settings stay in your own Docker volume. Open WebUI's docs give one Docker command to run it beside Ollama on the same machine, then open localhost:3000. Monthly cost: 0 USD in software; runs on hardware you own.
For you: A candidate home for Mantis's private wellbeing work; Sentinel reviews first.
Watch out: Open WebUI's docs say a stock install still makes some outbound calls; turn them off with ENABLE_VERSION_UPDATE_CHECK=false, ENABLE_OPENAI_API=false (or delete the default OpenAI connection) and RAG_EMBEDDING_MODEL_AUTO_UPDATE=false.
Checked 26 Oct 2026
A public customer-service Hermes bot on Telegram that answers from a knowledge folder, with a cheap free model for background work. Monthly cost: not stated.
For you: The main and auxiliary split is a direct cost saver for Chris.
A local agent built from plain markdown files, with mission control, a twice-daily security audit and Claude Cowork as the safer option for email. Monthly cost: not stated.
For you: Her twice-daily audit is a model for Sentinel.
Watch out: She avoids ClawHub skills unless the author is trusted.
The beginner Hermes shape, priced from public pages in late Oct 2026. Hostinger's Canadian page lists KVM 1 (1 vCPU, 4 GB RAM) at CAN$9.09 a month on promotion, paid up front for 24 months (CAN$218.16), renewing at CAN$16.79. Monthly cost: about CAN$9.09 hosting at the promo rate, plus model use in USD (Claude Haiku 5.5 is 0.10 USD in and 0.50 USD out per million tokens for prompts under 100,000 tokens; OpenRouter adds a 5.5% fee when you buy credits).
For you: Close to Chris's own shape. OpenRouter's logs showed all seven of his agents on Claude Haiku 5.5 on 26 Oct 2026, so he already has the cheapest rate in this list.
Watch out: Promo prices need a long up-front term; budget for the renewal price.
Checked 26 Oct 2026
Not a full stack: the model layer is usually the biggest monthly cost, so the price gaps matter. Anthropic's pricing page, per million tokens in USD: Claude Haiku 5.5 0.10 in and 0.50 out (prompts up to 100,000 tokens; 0.50 and 2.50 above that); Claude Haiku 4.5 1 in and 5 out; Claude Sonnet 5.5 2 in and 10 out. OpenRouter's Standard plan charges a 5.5% fee when you buy credits, not per request. Monthly cost: depends on use; these are rates, not totals.
For you: Efficiency win to check: if any agent still runs Haiku 4.5, Haiku 5.5 is listed at a tenth of the price; test it on that agent's real jobs first.
Watch out: Prices change; re-check the page before deciding.
Checked 26 Oct 2026
Hostinger's larger plans, for running agents plus n8n, Qdrant and monitoring on one server. Monthly cost (Hostinger Canada, promo, paid up front for 24 months): KVM 2 (2 vCPU, 8 GB) CAN$12.29, renewing at CAN$20.99; KVM 4 (4 vCPU, 16 GB) CAN$18.19, renewing at CAN$40.49; KVM 8 (8 vCPU, 32 GB) CAN$36.29, renewing at CAN$69.79. Model use is extra, in USD.
For you: The plan size Chris actually runs is in his Hostinger account; the renewal price is the number to plan around.
Watch out: Prices exclude tax, per the page; HST applies in Ontario.
Checked 26 Oct 2026
For testing ideas before spending anything. OpenRouter's free plan gives 25 or more free models with a limit of 50 requests a day. Tailscale's Personal plan is free for up to 6 users. Uptime Kuma is open source. Monthly cost: 0 USD, within those limits.
For you: Good for Sentinel to trial a new workflow without touching paid keys.
Watch out: Free models are often slower and weaker at tool use; never send private data to a provider you have not reviewed.
Checked 26 Oct 2026
From idea to a live site, in the order the videos do it: design, build, data and login, deploy, then what to run after launch. Coding habits that apply to every stage are in the 'Vibe coding' tab.
From your videos: 28 of the 105 videos feed this tab. Metics Media's Dokploy guide is the only full deploy in the videos, and none covers Coolify. 14 entries from the web fill the gaps (marked From the web): Coolify deploys, domains, HTTPS and backups, Vercel and Netlify basics, Supabase login and key safety, and a pre-launch security checklist.
Jason Lee typed /design and asked for a landing page that shows every feature, told Claude to treat it as its design resume, and gave it a reference site. Claude added scrolling and animated sections he had not asked for, built with code rather than video. He used the Higgsfield MCP to make the images.
For you: Chris asked for design mock-ups in Claude Design before anything is built, so this is step one for any of his sites.
Julian Goldie installed UI UX Pro Max into Codex by pasting its GitHub link and saying 'install this skill for website design'. He says you can either give a design skill or a screenshot of what you want; without either, pages look bland. Jack Roberts made his own skill, Power Design, with about 20 design rules such as the golden ratio.
Impeccable adds 24 design commands, such as critique and delight, to make a page less generic. Reimagine it is a redesign skill that first reads your files to learn the existing design system.
For you: A refresh path for the Gauntlet dashboard's look, checked in Claude Design before Forge ships it.
Nate Herk built Scrollcraft for Apple-style scrolling sites; it adds a scroll engine, design rules and a checker that screenshots the page so Claude fixes layout itself. 21st.dev is a library of interactive UI parts (components) Claude can pull through its MCP and CLI. The Shaders repo gives 200 free animated backgrounds.
Watch out: Scrollcraft is the presenter's own repo. The 21st.dev CLI needs a sign-in.
Both Jason Lee and Julian Goldie run the site on localhost, meaning only on your own machine, and click through it in Chrome before anything goes online. Julian notes Codex also has its own preview pane. You get the first 80% fast; the rest is feedback.
Alex Finn had Grok Bot build a micro-business site. Its built-in Cursor cloud agents made the first live version in about two minutes. He then connected Vercel for hosting, GitHub for the code, Resend for email, Stripe for payments and Google Search Console for search, and the bot asks his approval before outside actions.
Watch out: Alex Finn reports sponsorship revenue from his own large audience; Dubibubi's 10-day money test made 4.55 US dollars.
Anthropic's web artifacts builder skill turns a plain description into a small working page, such as a calculator or a dashboard. It suits one-page tools rather than full apps.
Julian Goldie asked his agent to let people log in with Supabase so each user's data is saved, and the agent added the login page. Nate Herk says Supabase's official skills teach Claude its real security rules, like making sure users only see their own data. Lead Gen Jay uses Supabase as a free back-end database for his agent.
Watch out: Nate Herk's read-only and test-user checks are the safety step; do not skip them.
On his own server, Metics Media created a PostgreSQL database in Dokploy, copied its internal connection URL, and added it to the web app as an environment variable (a setting stored outside the code) called DATABASE_URL. After a reload, the app's test page switched from 'db false' to 'db true'.
For you: Coolify offers the same kind of database service, so the steps carry over.
Nate Herk says to install only Stripe's 'best practices' skill, which teaches checks such as confirming a payment really came from Stripe. He warns that the full plugin may push every vendor choice through Stripe. Alex Finn's bot used a Stripe connector to make payment links.
Watch out: Keep payments behind an approval step; agents should not spend or charge on their own.
Jack Roberts says never paste an API key (a secret password for a service) into a chat; set it as an environment variable. Lead Gen Jay keeps all keys in Doppler, a central secrets store that syncs to Vercel. Sharbel A. warns not to put secrets on a Vercel dashboard site.
For you: The same rule as Chris's no-secrets-in-Discord rule.
Supabase Auth handles sign-in by email and password, magic link, one-time codes, 19 listed social logins such as Google and Apple (plus any OAuth or OIDC provider), and SSO. It stores users in your Postgres database. The login token limits each request to the rows a user may see, but only through the RLS rules you write.
Watch out: Login alone does not protect data. Without RLS rules, a logged-in user may see everything.
Checked 26 Oct 2026
Supabase says to enable Row Level Security (RLS, rules for who sees which rows) on every table the API can reach. Without it, the table is open. Once on, no data comes through the public key until you add policies.
Watch out: The 2025 Lovable data leak came from weak RLS rules (see the 'Vibe coding' tab).
Checked 26 Oct 2026
Supabase's publishable key is safe in web pages and apps because RLS limits it. The secret key bypasses RLS and gives full access, so it belongs only in server code, never in the browser, a shipped app or git. The old JWT-based anon and service_role keys are legacy; Supabase recommends moving to the new keys (the end-of-2026 phase-out date could not be confirmed on the current page).
Checked 26 Oct 2026
Alex Finn calls Vercel his favourite host and says you can host a small site completely free. Sharbel A. hosts his agent dashboard on Vercel, gated behind a Google login so only he can see it. A Vercel plugin also appears in AI Edge's design-agent demo.
Watch out: Sharbel A. says a dashboard on Vercel must not expose private files or secrets, and risky actions should need approval.
Julian Goldie set up a Codex cloud routine (a scheduled job that runs with the app closed) that writes five SEO articles a day for each of his five Netlify sites and publishes them through the Netlify connector. He says cloud routines suit API work like this, not tasks that need to click around a screen.
Watch out: He says a new routine needs feedback in its first few days before it runs well.
Metics Media deploys a web app to a Hostinger KVM 2 VPS using the Dokploy template, which comes with Docker and Dokploy already installed. Dokploy pulls the code from GitHub, builds it with Nixpacks (which detects Node, Python or static sites by itself) and runs it. It is the same kind of panel as Chris's Coolify.
For you: Coolify follows the same flow: GitHub app, build, domain, certificate. Remember Chris's Coolify rule of Stop then Deploy for changes.
Watch out: Lead Gen Jay calls Hostinger's servers weak and runs his agent on Hetzner instead.
Metics Media's beginner guide hardens a fresh VPS before anything else. It adds a non-root user with sudo rights, logs in with an ed25519 SSH key instead of a password, and turns on the UFW firewall allowing only SSH and web ports 80 and 443.
For you: Sentinel can check Chris's VPS against this list.
Metics Media deletes the default DNS records, then adds A records (address records) pointing to the VPS IP for @, www and subdomains such as admin and analytics, with a TTL of 300 so changes spread fast. In Dokploy he adds the domain, sets the app's port (3000 for Node.js), picks Let's Encrypt for a free HTTPS certificate and presses Validate DNS. He also moves the admin panel itself onto a domain, because the raw IP showed 'Not secure'.
For you: Chris's agent86.cloud subdomains follow this pattern already.
Watch out: Cloud X Berry's DNS Globe checks that a DNS change has spread; Switch and Click uses Cloudflare and Nginx Proxy Manager for home servers.
With Autodeploy on, each commit to GitHub makes Dokploy rebuild the live site with no extra step. Preview Deployments give each pull request (proposed change) its own test link, using a wildcard domain, so you can check a change before it goes live.
Julian Goldie says agent dashboards belong behind a VPN or SSH tunnel. Nimrod and Pav Rusovs use Tailscale, a private network linking only your own devices. Tina Huang uses Cloudflare Tunnel only for demos, and bashbunni uses Tailscale Serve to share a local port with her own devices.
For you: Applies to the Gauntlet and Hermes dashboards.
Tina Huang lists Railway, Vercel, Hostinger and Heroku for deploying apps, and Hetzner and Hostinger for VPS servers at about 5 US dollars a month. Lead Gen Jay drives Hetzner from Claude Code with the Hetzner CLI, which can create servers.
Watch out: Letting an agent create servers can start paid subscriptions; keep an approval step.
Nate Herk runs Claude Security, Anthropic's plugin that sends a team of agents through the code looking for holes, with other agents trying to prove each finding wrong; you get a plain-English report and patches. Jack Roberts names a Cloudflare security audit skill that turns a coding agent into an auditor.
For you: A natural job for Sentinel, who reviews but never installs.
Coolify connects to GitHub through a GitHub App, which gives it access only to the repos you pick. You then create an application from that repo, set the branch, build pack and port, add the domain and environment variables, and deploy.
For you: Chris already runs Coolify (coolify.agent86.cloud), so this is the home route for his own web apps.
With the GitHub App, Coolify switches on Auto Deploy for you, so each push redeploys the app. Without it, you set a webhook in GitHub with a shared secret. Coolify can also build a preview for each pull request.
Watch out: The webhook secret works like a password; keep it out of Discord and the repo.
Create a DNS A record that points the name to the server running the app. In Coolify, add the domain with https:// and redeploy; the built-in proxy gets and renews a free Let's Encrypt certificate. Ports 80 and 443 must reach the server.
Watch out: Generated sslip.io addresses are for testing only; Coolify says to use your own domain, or a wildcard you own, when you need HTTPS.
Checked 26 Oct 2026
Click New Project in Vercel, pick the Git repo, check the framework preset, root folder and environment variables, then Deploy. After that, every push gets its own preview URL, and merging into main makes a production deploy. You can roll back by reverting.
Watch out: The free Hobby plan cannot deploy a private repo owned by a GitHub organization (or GitLab group or Bitbucket workspace), and only the Hobby team owner's own commits deploy.
Checked 26 Oct 2026
In Netlify, choose Add new project > Import an existing project, connect GitHub and accept the detected build settings. Each pull request gets a Deploy Preview at a deploy-preview-<number> address. Previews can be password-protected and are hidden from search engines.
OWASP's Top 10 (2025) is the standard list of the worst web app risks. Number one is broken access control, meaning users reaching data that is not theirs. Others include security misconfiguration, supply-chain failures, injection and authentication failures.
For you: A ready-made checklist Sentinel can work through.
Checked 26 Oct 2026
Mozilla's HTTP Observatory checks a site's HTTP headers, such as CSP and HSTS (rules that tell browsers how to protect visitors), and gives specific fixes.
GitHub's push protection stops a push that contains a key or token. For your own account it is on by default for public repos. Repo-level push protection is off by default and needs GitHub Secret Protection.
For you: Fits the no-secrets rule, since the vault and repos sync to GitHub.
Checked 26 Oct 2026
Pulled from the official pages above. Work through it in order, and do not go live until each line passes.
For you: Sentinel reviews; Chris gives the final yes.
Metics Media adds Umami, a free analytics tool, from a Dokploy template on its own subdomain with HTTPS. He adds the site in Umami, copies the tracking code into the page's head section, commits it on GitHub, and auto-deploy does the rest. Page views then show up.
Watch out: The Umami template ships with the default login admin / umami. Change it on day one.
Metics Media switches on Hostinger's weekly snapshots and backups for the VPS, under Backups and Monitoring. Daily backups are a paid upgrade.
For you: Check the backup schedule for Chris's Coolify VPS.
NetworkChuck uses Checkmk's free Community Edition to watch websites as well as servers and containers. Its HTTP check confirms a site is up, returns the right status and shows expected text, and it can send alerts to chat.
For you: An alert like this could post to Sentinel's Discord channel, with no secrets in the message.
Nate Herk uses Open SEO for real keyword, ranking and backlink data, so Claude can name the exact page and keyword to fix. Alex Finn's bot asked to connect Google Search Console and tuned the site so chatbots recommend it. Dubibubi lists an SEO audit skill with 12 sub-skills.
Watch out: Open SEO's free trial credits are small, and your data sits on their server unless you host it yourself.
Alex Finn uses Resend, a programmatic email service, so his agent collects sign-ups and sends a daily newsletter, tracking opens and clicks. The bot also places sponsors in the newsletter.
Watch out: Other presenters keep agents to drafts only for email; sending on its own is a choice to make on purpose.
Coolify can back up PostgreSQL, MySQL, MariaDB, MongoDB, ClickHouse and SQLite on a schedule, to the server and to S3 storage (offsite cloud storage). Coolify's docs stress that a backup only proves a file exists; restore one into a spare database to prove it works.
Watch out: Redis, Dragonfly and KeyDB are not covered; they need their own backups.
Checked 26 Oct 2026
A Coolify instance backup saves projects and settings, not app data. Coolify's APP_KEY encrypts saved secrets, and it is not in the backup (S3 instance backups hold only the database file). Without it, restored secrets cannot be read.
Watch out: The APP_KEY is a secret: never post it in Discord.
Checked 26 Oct 2026
What the videos show about building phone, tablet and desktop apps with AI: design first, the skills that help, and the steps to reach the App Store. Use it to plan Cyborg's Swift work.
From your videos: 15 of the 105 videos touch native or mobile apps, and only one (Jason Lee) builds an iPhone app on screen. No video covers iPadOS, a native Mac build, Android Studio or Google Play, so 16 entries from the web fill those gaps (marked From the web): Xcode's AI and agents, Swift skills, iPad rules, TestFlight, App Store and Play Store steps, Mac notarization and Android Studio's Gemini.
Android Studio has a built-in agent. You describe a goal, and it plans, edits several files, builds and fixes until done. It waits for you to accept or reject each change. It can also run the app on a device, take screenshots and read logs.
Google publishes official Android skills that work with any tool that supports the agent skills standard. There are 24, including 'Use ML Kit GenAI Prompt API', which sends requests to Gemini Nano on the phone itself. This is the 'Android skills' pack the videos name.
Watch out: Running android skills add on a skill you already have updates it to the latest version, which can replace your own edits; rename a skill before you change it.
Checked 26 Oct 2026
Personal Play Console accounts made after 13 Nov 2023 must run a closed test before going public. You need at least 12 testers who stay opted in for 14 days in a row. Then you apply for production access, and Google usually replies within 7 days.
Watch out: A tester who leaves and rejoins restarts their 14 days.
Checked 26 Oct 2026
The Next New Thing lists an official Android skills pack. It includes skills for on-device AI, meaning models that run on the phone itself, such as Gemini Nano and ML Kit. This is the only Android build tool named in the videos.
For you: Only relevant if Chris wants an Android version. Cyborg works in Swift today.
Watch out: The repo path was not given, and no video shows the Google Play Store steps.
Since August 2021, new Play apps must be uploaded as an Android App Bundle (Google's upload format). Before release, turn off debug mode and logging, and test on at least one phone and one tablet. In Play Console you create the app, fill in the store listing, then upload the bundle.
Watch out: Store listing limits: name 30 characters, short description 80, full description 4,000.
Checked 26 Oct 2026
Android Studio can use other AI providers, not just Gemini. Google's page lists OpenAI, Claude and OpenRouter as examples. Chat and agent features work, but some special Android Studio AI features may not.
For you: Chris already pays for models through OpenRouter, so no new account is needed.
Watch out: Your code goes to that provider. Never put the API key in source code.
Checked 26 Oct 2026
Julian Goldie points out that his first habit-tracker app had no login and stored no data, so it could not go public. He asked the agent to add Supabase, a free database and login service, so each user gets an account and their own saved data. The agent then logged in to Supabase and added the login page.
For you: Any Cyborg app that stores user data needs this layer. The security checks for it are in the 'Build and deploy web apps' tab.
Watch out: Nate Herk says to set the Supabase connector to read-only for the agent.
Every TestFlight or App Store release starts as an archive, which is a packaged build. Xcode's Organizer can check it, then upload it to App Store Connect. If the app has no App Store Connect record yet, Xcode asks for the details and makes one.
Watch out: You need a paid Apple Developer Program membership to upload.
TestFlight is Apple's beta-testing app. You can invite up to 100 internal testers (your App Store Connect team) and up to 10,000 external testers by email or public link. The first build added to an external group goes through App Review. Each build works for 90 days.
Checked 26 Oct 2026
In a fresh chat, Jason Lee typed /design again and asked for a landing page that shows all the app's features. He reused the same reference so the site matched the app, and asked Claude to use the Higgsfield MCP (a link to image and video models) for pictures. He says the result did not look vibe coded.
For you: One design language for the app and its site keeps Chris's products consistent.
If you run Claude Code or Codex in Terminal, Xcode can lend it its tools, such as building your app. This works through MCP (a standard plug for AI tools), using Xcode's own bridge. Xcode shows an alert when an outside agent connects.
For you: Lets Chris's Mac Claude Code build and test Swift projects the same way Xcode's own agent does.
Checked 26 Oct 2026
To clone an invoice app said to earn about 2 million dollars a month, Jason Lee gave Claude the App Store link and the app's own website. Claude read both and listed every feature for his version. He chose that app because contractors on the go need quick invoices from a phone.
For you: Cyborg could use this to scope one of Chris's own products, such as the submit-app, against rivals.
Apple's design guide says iPad users sit about 3 feet away, often run several apps at once, and mix touch, keyboard, trackpad and Apple Pencil. Use the big screen for content, avoid full-screen pop-ups, and adapt to rotation, multitasking, Dark Mode and text size changes.
Nate Herk says Expo's official skills take you from an idea to an app running on your own iPhone in an afternoon, even from Windows. They include Expo's list of 20 things that make an app look AI-built, plus the current rules for getting into the App Store. He says to name the Expo skills in your prompt, because they do not always load on their own.
For you: The video does not say these skills write Swift. Check with Cyborg whether they suit a Swift build or only an Expo one.
Nate Herk says you need a 99 US dollar a year Apple developer account for TestFlight (Apple's beta-testing app) or the App Store. Julian Goldie's course shows the next step in plain words: ask the agent to get the app ready to export to the iOS App Store, and it prepares the project for Xcode, Apple's app-building program.
For you: Cyborg already works in Swift, so the Xcode step is home ground. The account fee is the new cost.
Watch out: Neither video walks through Apple's review screens, so plan for that step without a guide.
Jason Lee says that without a reference, Claude falls back to the same colours and buttons it gives everyone. He used referral.design, a site that lists the design systems of known apps such as Wise, Spotify and Mercury, with exact colour codes and positions. He pasted the page link and Claude read it by itself.
For you: A good first step for any Cyborg app, because it fixes the look before code exists.
Watch out: Do not ship another brand's colours. Jason Lee kept the layout and asked Claude for new colour themes, because people would spot a copy.
Since iPadOS 26, the old setting that kept an app full screen (UIRequiresFullScreen) is deprecated. From iOS 27 and iPadOS 27, for apps built with the iOS 27 SDK, it no longer stops resizing, and a launch screen is required for App Store submission. Apple asks apps to support all orientations and adapt when the window changes size.
For you: Give Cyborg this note as a rule for any iPad build.
Watch out: Apple warns that apps that do not update may show broken or misplaced layouts.
Checked 26 Oct 2026
AI Edge lists gstack, Garry Tan's set of role skills, and names its commands. One of them is iOS QA, for checking an iPhone app. Others are /office-hours to question the idea and /codex for a second opinion.
For you: A possible test step for Cyborg after a build. The video names the command but does not demo it.
Watch out: Only the command name was shown, so check what it does before relying on it.
The /brag skill, free on GitHub, turns a project folder into a 20-second launch video built from code. Jason Lee ran it in the Quicksand folder, so it pulled the app screens and the website. Pasting a reference launch video improved the style, and asking Claude to list the video scene by scene made it easy to edit.
For you: Forge could make launch clips for Chris's products this way.
Watch out: The GitHub repo was not named aloud. One frame had a card covering a word, so check every frame.
Jason Lee typed /design in the Claude desktop app and asked for an invoice app called Quicksand. One prompt gave 22 screens on a canvas, grouped into onboarding, create invoice and preview. He then asked Claude to serve the mock-up on localhost (your own computer) so he could click every button and check that the flow worked.
For you: Chris likes to see a design before anything is built, so Cyborg can show Claude Design screens for sign-off before writing Swift.
Watch out: Jason Lee designed the screens only. He did not build the back end (the server and data side) in the video.
Julian Goldie keeps one project in Codex with several chats side by side: one for the look, one for adding plugins, and one for getting it ready for Xcode. Each chat works on its own part, front end or back end. He gives feedback in each chat and iterates.
For you: Maps onto Chris's 'one thing at a time' rule: each chat owns one job.
After beta testing, you submit the version in App Store Connect. Fill in the required details first, such as screenshots, privacy and age rating, and pick the right build. Apple does not always review submissions in the order they arrive.
Watch out: Only the Account Holder, Admin or App Manager role can submit. Some details cannot be changed after release, so choose carefully.
Checked 26 Oct 2026
Paul Hudson's SwiftUI Pro skill targets 'the mistakes LLMs actually make' in SwiftUI: navigation, layout, state, accessibility and deprecated code. He also keeps Swift Agent Skills, a list of community skills for SwiftData, Swift Concurrency and Swift Testing. This is likely the SwiftUI skill the videos mention, but the videos did not name the owner.
For you: The most direct match to Cyborg's Swift lane.
Watch out: The list's own README says being listed is not an endorsement; read each skill and trust the author before installing.
The Next New Thing hosts say Apple's Swift documentation is often out of date, so agents trained on it try and fail. This skill gives the agent current Swift and SwiftUI (Apple's screen-building framework) documentation so its code works with today's iOS. It is the most direct Swift item in all the videos.
For you: The closest match to Cyborg's Swift lane. Ask Sentinel to review it before install.
Watch out: No owner or repo was given in the video, so find and check the source first.
Xcode has a coding assistant built in. You switch it on in Xcode's Intelligence settings. You can sign in to Claude or ChatGPT for chat, and install coding agents such as Claude Agent and Codex. Xcode can also use any other chat provider that speaks the Chat Completions API (a common AI web interface).
For you: Xcode runs on Chris's Mac, so this is for Mac-side Swift work; Cyborg on the VPS cannot use it directly.
Watch out: Apple says the agent or model may read your project files. Check what each provider's terms allow first.
Xcode 26.3 (February 2026) let agents search docs, explore files, change project settings, build, fix and check their work by capturing Xcode Previews. Xcode 27 (WWDC, June 2026) added Google's agent next to Anthropic's and OpenAI's, a planning chat, and plug-ins for extra skills, with GitHub and Figma first. Type / in the prompt box to see Xcode's built-in skills, such as /plan for plan mode.
For you: Plug-ins add MCP servers and skills, so each one goes through Sentinel first.
Watch out: Agent settings for Xcode live in their own folder (~/Library/Developer/Xcode/CodingAssistant) and do not affect the same agent outside Xcode.
No video shows a native Mac app being built with AI. The nearest items are a mention of someone who vibe coded a Mac status-bar app for his air conditioner by taking apart the maker's app, and a test where Claude copied the Mac Preview app. In that test the reverse-engineering tool added little, because the plain model managed alone.
For you: If Cyborg builds a Mac app, the Swift skill and design steps above still apply.
Watch out: The hosts discuss where the legal and ethical line sits. Copying a paid app to give away was the line they would not cross.
Several videos show desktop apps built so AI can control them. Clop, a free Mac app that shrinks and converts files, exposes an MCP (a plug-in link for AI) for Claude. The Artcraft suite of free Adobe-style apps ships with a built-in MCP. CLI-Anything builds a command-line tool from any desktop app so agents can run it.
For you: A design idea for Cyborg: give Chris's own Mac tools an MCP so the Hermes agents can use them.
Watch out: These are examples of the pattern, not build guides.
A Mac app you give out yourself must be signed with a Developer ID certificate and notarized, meaning Apple scans it and stamps it. Gatekeeper, the Mac's install guard, checks for this. Xcode's Distribute App has a 'Direct Distribution' option that notarizes for you.
Watch out: Only the Account Holder can create a Developer ID certificate. For a Mac Catalyst app, make separate iPad and Mac archives.
Apple's Foundation Models framework lets a Swift app use the on-device Apple Intelligence model, or Private Cloud Compute for harder jobs. It can return Swift data types directly (@Generable) and call your own tools. Xcode 27 also lets the framework use other providers' models, including Claude and Gemini.
For you: On-device AI keeps data on Chris's devices, which suits private features.
Watch out: The on-device model needs a device that supports Apple Intelligence.
Brendan Jowett used GPT-6 Astra in the Codex desktop app to build clones of Mario, Subway Surfers, Counter-Strike, Minecraft and GTA in Unreal Engine 5, a free game engine. The agent drove Blender, a free 3D tool, through an MCP to make models. Julian Goldie also built a 3D racing game through the Blender MCP.
For you: Matches Cyborg's sim-racing interest, for example a small racing prototype.
Watch out: In the Mario test the model found three Mario clone repos on GitHub and used them, so Brendan re-ran it without open-source assets. One-prompt demos may lean on existing code.
Greg Isenberg's guest gives agents the Mobbin MCP, which opens Mobbin's library of real app screens and user flows. The agent looks at how working apps solve a screen before it designs one.
For you: Useful before a Claude Design mock-up, for Chris's visual sign-off.
Tina Huang lists Apple Intelligence and Gemini Nano as edge models, meaning AI that runs on the device rather than in the cloud. They suit features that must work offline or keep data on the phone.
For you: Fits Mantis-style private features, where data should stay on Chris's devices.
Watch out: Only named in passing, with no build steps.
REA is a fast-growing repo of skills and MCP tools that lets an agent study how an app works and rebuild its features. Presenters claim between 240,000 and 300,000 GitHub stars. A demo had the agent inspect the game 2048 and build a one-file working copy.
For you: Sentinel should review it first. It is a research tool, not something to point at paid apps.
Watch out: The Next New Thing hosts note that software terms often forbid reverse engineering, and say copying a paid app to give away crosses the line.
Tools that work best together, as the presenters show them: the goal, the parts, how they connect, and which agent it suits.
From your videos: Draws on 67 of the 105 videos, starting from the dashboard combos list. Where a combo joins tools no single video showed together, the summary says who showed each part. Combos the warnings list flags (Scrapling, Agent Reach, giant skill packs) were changed or noted. Plus 9 entries from the web, from official docs (marked From the web): Hermes webhooks, n8n with Discord, MCP and OpenRouter, Obsidian over MCP, Coolify with Tailscale, Uptime Kuma, and Home Assistant. Updated 26 Oct 2026: 4 combos from full transcripts of x2F3fdn82GU, IbFaY3xFpZM and BrBQRNIDRS8.
Goal: A refreshed dashboard look based on real app screens. Real screens ground the design brief, and each skill covers one step.
For you: Cyborg for the Gauntlet code; Forge for any images.
Watch out: Third-party skills; Sentinel vets them.
Goal: A feature built from questions to reviewed code, with the plan saved in files. Questions come first, and every stage leaves a file, so a reset loses nothing.
For you: Cyborg (code and iOS).
Watch out: Install one stage at a time; many skills bloat context.
Goal: The same coding work at a smaller token bill. Measure first, then trim.
For you: Cyborg.
Watch out: Nate Herk measured no saving from RTK; changing MCPs mid-session resets the cache.
Goal: Fix and update a remote agent without logging in by hand. Lead Gen Jay runs a custom /bob command in Claude Code that SSHes into his Hermes server and edits it, with tokens kept in Doppler and the server made with the Hetzner CLI.
For you: Cyborg; Chris uses Coolify instead of Hetzner.
Goal: A long-running build that keeps going until it meets a clear goal. Alex Finn has Hermes write a detailed /goal prompt first, answers its questions, then lets it run for hours; he finishes the code in Claude Code or Codex.
For you: Cyborg.
Watch out: Long runs spend tokens; set the turn cap.
Goal: Several coding tasks side by side without clashes. Worktrees keep changes apart and /goal keeps each agent going until its test passes.
For you: Cyborg.
Watch out: Parallel agents multiply spend.
Goal: Check new skills and risky actions before they bite. Tools do the first pass and Sentinel does the final read.
For you: Sentinel reviews; Cyborg installs only after Chris's yes.
Watch out: Scanners miss things; a human still reads the file.
Goal: Hand a coding job to the right tool and get a report back. Tina Huang's Coder agent takes a spec in Discord, hands building to Claude Code, and a local Qwen model reports back. AI LABS shows Hermes delegation skills for Claude Code, Codex and Cursor.
For you: Cyborg.
Goal: Replies drafted in Chris's style, never sent automatically. Sharbel A. keeps inbox replies draft-only. Dan Martell builds a voice style guide from 20 recent emails so drafts sound like you.
For you: Alfred; Jack Roberts says do not let an agent send email.
Goal: Action items from each meeting become tasks without retyping. Sharbel A. has Hermes pull Fireflies transcripts and assign action items as Notion tasks. Tina Huang's agents read Granola notes through the Granola MCP.
For you: Alfred, filing into the vault instead of Notion.
Goal: A customer-facing chat bot that answers from your own documents. Python Simplified gives Hermes a persona, an agent-knowledge folder of five markdown docs, a public Telegram bot with admin commands limited to one ID, and a daily report cron that runs a Python script first.
For you: Not a current need; a pattern for the submit-app later.
Watch out: One environment setting opens the bot to everyone; check it.
Goal: A short audio overview of the day instead of a wall of text. Tina Huang has Hermes use a NotebookLM MCP to make a podcast-style daily brief. Gemini Notebook (the renamed NotebookLM) makes Audio Overviews from sources.
For you: Alfred; feed only a public-safe folder, never the archive or Mantis's material.
Watch out: This sends notes to Google.
Goal: Phone voice notes become tidy vault notes. Nimrod sends voice notes to Hermes from an iPhone Action-button shortcut, transcribed by Whisper, into a wiki inbox. The dashboard notes sketch an n8n version.
For you: Alfred captures, Brainiac files.
Watch out: Keep secrets out of voice notes; transcripts land in Discord and the vault.
Goal: A palette, logo ideas, social graphics and hero art. Style rules are written once, so every later image follows them. Jason Lee anchored Claude's design with referral.design.
For you: Forge.
Watch out: GPT Image 2 is paid; set a cap before the 01 Nov 2026 reset.
Goal: a design you can still edit by hand. AI Samson asks Claude, through Artcraft's MCP, to find one of his YouTube thumbnails and rebuild it in PhotoCraft with background, subject, text, logos and effects on clearly named layers. He then changes any part himself.
For you: Forge (photo and design).
Watch out: His logo-tracing test in VectorCraft fell short. The video's Nano Banana 2.1 and Higgsfield part was a paid sponsor segment.
Goal: A printable poster of the seven agents and how they connect. Diagram tools keep text accurate; the image model only sets the mood.
For you: Forge.
Watch out: Never let an image model draw the labels.
Goal: Images made on Chris's own Mac at no per-image cost. Local generation is the cheapest option. Tina Huang and Better Creating both show Draw Things.
For you: Forge.
Watch out: Local quality is lower and depends on the Mac's memory.
Goal: A saved scorecard of which image model suits which job. A fixed prompt set stops opinion from deciding.
For you: Forge.
Watch out: Check current model names first; captions garbled some.
Goal: A no-code agent on open models with no API bill. Tina Huang uses n8n's self-hosted AI starter kit, which bundles n8n, Ollama, Qdrant and Postgres.
For you: Brainiac for the vault index idea; useful as a test bench.
Goal: Free, private ingestion of sources into the vault. Wanderloots drafts wiki notes with Ollama and uses double-click launcher commands that put drafts in a human review folder first.
For you: Brainiac.
Goal: Speak to an agent and hear it reply without cloud voice fees. Tina Huang chains Whisper (speech to text), a small Qwen model and Piper (text to speech) even on a Raspberry Pi. Komputer Mechanic found Piper about twice as fast as Kokoro.
For you: Alfred, as a free fallback to ElevenLabs.
Goal: Ask questions of your own files with nothing leaving the machine. Better Creating runs Ollama with AnythingLLM for document upload and embedding, and Open WebUI, which plugs into n8n and MCP. He reaches the Mac mini from an iPad with Screens 5.
For you: A candidate for Mantis's private wellbeing work, since nothing goes to the cloud; Sentinel reviews first.
Watch out: Better Creating says Wispr Flow stopped working offline.
Goal: Talk to the house with no cloud model. Home Assistant's Ollama integration adds a conversation agent that runs on your own Ollama server. It can control devices you expose, but only with models that support tools.
For you: Keeps home data off the VPS; Mantis-style privacy for the house.
Watch out: Home Assistant's docs call device control experimental and say smaller models make more mistakes.
Goal: A technical report on how another app works. Alex Finn asks Hermes to open a browser, click around a competitor's site, read the console and report the stack, features and pricing, then hands the markdown to Claude Code or Codex.
For you: Cerebra researches; Cyborg builds if needed.
Watch out: Approve each browsing permission it asks for.
Goal: A weekly cited brief on a topic, filed in the vault. Gathering, citing and writing are separate steps, so a thin source shows early.
For you: Cerebra.
Watch out: Reddit and X are opinions; label them. Agent Reach was dropped because the warnings list flags bot-block bypass tools.
Goal: A growing linked wiki where each source becomes a note. Sources are compiled once into linked notes instead of re-read each time.
For you: Brainiac files; Cerebra queries.
Watch out: Keep the archive folder out of it.
Goal: Turn web pages into clean text cheaply. Defuddle strips a page to reader-mode text to save tokens; Crawl4AI and Firecrawl turn sites into markdown; Apify MCP scrapers handle bulk data. The deep research skill can then split work across models.
For you: Cerebra.
Watch out: Nuno Tavares says skip Scrapling because it gets around site blockers; it was left out. Respect site terms.
Goal: A structured report with sources kept per section. AI Foundations splits research into stages (refine, manage, research, store) so each can be checked.
For you: Cerebra; fits Chris's existing n8n and OpenRouter.
Watch out: Check every citation matches a real source before filing.
Goal: A spoken research summary emailed when a question is submitted. Tina Huang's no-code build follows her six agent parts: model, tools, memory, audio, guardrails, orchestration.
For you: Cerebra.
Watch out: Sonar and audio calls cost per use; cap daily runs.
Goal: A searchable index of the vault agents can query. It matches Chris's Qdrant plan and the n8n starter kit Tina Huang showed.
For you: Brainiac.
Watch out: Test with and without Qdrant on the same questions first; Mason Fitzgerald's numbers are his own test.
Goal: Lessons from a pile of videos turned into notes and skills. This is how Chris's 105 videos could become a searchable library.
For you: Cerebra.
Watch out: Auto-captions garble names; verify before installing anything.
Goal: Know when an agent's gateway, crons or model login fail, and get a fix ticket. NetworkChuck runs Checkmk's free Community Edition with a small local script on his Hermes VM that reports crons, gateway, model login and usage. Alerts go to Slack, and a second rule opens a Paperclip task so an agent restarts the gateway.
For you: Sentinel watches and reports; the fix stays with Chris's approval. Alerts could go to Sentinel's Discord channel.
Watch out: The video is sponsored by Checkmk; the free edition needs manual plugin setup for Docker and other extras.
Goal: A small team where one lead bot takes requests and hands work to specialists. Sharbel A. builds an orchestrator, a researcher and a creator in Hermes bot mode. The lead runs on a strong model, workers on cheap or local ones, and each worker gets only the tools its job needs.
For you: Alfred as dispatcher, with Cerebra and Brainiac as workers; Mantis stays out of shared rooms.
Watch out: Sharbel A. says some tools load up to 50,000 tokens into every message; a one-off job does not need a team.
Goal: Cut model spend on routine jobs. Python Simplified sets a cheap auxiliary model for background tasks and ticks the no-agent box on script-only cron jobs. Jack Roberts and Metics Media add per-key spend limits in OpenRouter.
For you: Fleet-wide; fits the 01 Nov 2026 budget reset.
Goal: Keep each agent lean and its skills clean. Jack Roberts shows Bouncer, which scans a new skill for prompt injection and then allows, asks or quarantines it, and Context Doctor, which audits soul, memory, skill descriptions and tools for bloat and quarantines rather than deletes.
For you: Sentinel reviews; quarantine-not-delete suits Chris's archive rule.
Watch out: Both are third-party skills, so they need Sentinel's review before install.
Goal: Let outside services wake an agent and post the result to Discord. Hermes has a webhook adapter: it accepts signed POST requests, turns each one into a prompt, and can deliver the reply to a Discord channel. GitHub signatures are supported directly.
For you: Sentinel could get GitHub alerts for the vault repo in her channel.
Watch out: Hermes docs: a valid signature proves who sent it, not that the text is safe; treat PR and issue text as untrusted.
Goal: Start each day with every agent-sized task on one board. Alex Finn writes his to-do list, then drops anything Hermes can do into the Kanban triage column, and Hermes auto-assigns it to agents. Sharbel A. adds a 09:00 cron that asks for the day's number-one priority.
For you: Alfred: briefings and reminders are his lane; he can post the board summary to his Discord channel.
Goal: Workflow results land in the right agent's Discord channel. n8n's Discord node can send messages, wait for a reply, react and manage channels. A webhook credential only posts to one channel and cannot listen; a bot credential can do both.
For you: Fits the existing n8n nightly note: post the draft to Alfred's or Brainiac's channel.
Watch out: Keep tokens and webhook URLs in n8n credentials, never in a Discord message.
Goal: Let an agent fetch files or test work across Chris's devices without opening ports. Alex Finn installs Tailscale on all devices so Hermes can work across them, fetch a file to his phone or test a local web app on another machine. He calls it his number-one Hermes use.
For you: Cyborg for admin jobs; Sentinel checks which devices an agent may reach.
Watch out: Giving an agent SSH to every device is broad access; scope it.
Goal: Recover quickly when an agent breaks. Alex Finn runs at least two Hermes profiles so one can repair the other. Samin Yasar uses Codex as the doctor that fixes Hermes; Alex Finn opens the Hermes folder in Claude Code or Codex.
For you: Cyborg (code) can be the repair agent for the rest of the fleet.
Goal: skills that stay small and safe. Dubibubi's rule: a good Hermes skill stays under 15 KB or it bloats memory and burns tokens. SkillClaw can merge and rewrite skills after each session; the warnings list says to scan any skill for hidden instructions first.
For you: Sentinel and Brainiac.
Watch out: An auto-rewriting tool must not bypass Sentinel's review.
Goal: Make sure no agent spends, sends to a client or deletes without a yes. Sharbel A. lists three places to put a gate. Removing the tool is the strongest, the Hermes approval setting is next, and a line in the soul file is the softest.
For you: Sentinel: matches Chris's rule that nothing is installed without Sentinel's review and Chris's yes.
Goal: Know when a service on the VPS goes down. Uptime Kuma is a free, self-hosted monitor that checks websites, ports, ping and Docker containers, and can alert Discord among over 90 services.
For you: A plain, script-free health check that sits beside the agent health monitor; alerts could go to Sentinel's channel.
Goal: A short video in the structure of a viral one, on Chris's own topic. Presenters point to Hypit for borrowing a format, not a script.
For you: Forge.
Watch out: Nuno Tavares says skip Hypit because it needs paid services; copy structure only, and check voice-cloning consent.
Goal: An animated explainer that uses generated images or clips. Tristen O'Brien adds Higgsfield to Claude as a custom connector, builds the animation in Claude Motion, and uses Remotion through Claude Code as the free route.
For you: Forge.
Watch out: Claude Motion is a beta for Team and Enterprise plans only, per the video.
Goal: an animation you can tweak instead of regenerate. AI Samson had Claude build an animation inside EffectCraft (the After Effects clone), then opened the project to change timing, delete layers and swap a word. His reason: an AI video model gives a finished clip, this keeps full control.
Watch out: EffectCraft can run After Effects scripts but cannot open .aep project files, has no plugins and only basic 3D.
Goal: A short promo clip from a plain product or app photo. Each tool does one narrow job, and a person approves before posting.
For you: Forge.
Watch out: Pika runs inside ChatGPT, so there is a manual hand-off.
Goal: A 20-second launch video made from the project itself. Everything is code, so a text change is a re-render. Jason Lee showed /brag; Remotion gives fine control.
For you: Forge (video editing).
Watch out: Jason Lee did not name the /brag repo, so Sentinel vets it first.
Goal: A talking-head clip from one image and one audio track. A still plus audio is far cheaper than full video generation.
For you: Forge.
Watch out: Names came from garbled captions; confirm each model and price first.
Goal: A 3-minute weekly AI news video with sources on screen. Research and production are separate, so a weak script is caught before the render.
For you: Cerebra researches, Forge produces.
Watch out: Open Montage (700+ skill files) was left out per the giant-skill-pack warning.
Goal: Reach Coolify and admin ports only over Tailscale. Coolify offers a Tailscale Client service so a deployment can join your tailnet. Coolify's docs say that if you lock down port 22, allow SSH from the Tailscale range (100.x.x.x/10), because Coolify needs SSH to deploy. Tailscale's Personal plan is free for up to 6 users with unlimited devices.
For you: Extends 'Reach every device privately' to the VPS itself; Sentinel reviews firewall changes before they go in.
Watch out: Changing SSH rules can lock you out; keep a console login from the host's panel ready.
Goal: Let an AI client check and control the house. Home Assistant (since 2025.2) has a Model Context Protocol Server integration. You copy the server URL from Settings > System > AI into the client and sign in with OAuth. Clients can only touch the devices you expose.
For you: Alfred could answer 'is the garage door shut?' in his channel, with only a few devices exposed.
Watch out: Cloud-hosted agents need Home Assistant reachable from the internet, for example through Home Assistant Cloud; that widens exposure.
Goal: lower AI cost and more dependable jobs. After hitting 96% of his weekly Grok allowance, one host had his Zapier agent go through his team's work and find jobs that could become zaps. Zaps wait for webhooks (a ping when something happens) and run code, so they are cheaper and steadier than an agent checking constantly. Anything needing judgment is handed back to the agent.
For you: Chris already runs n8n, so the same move works there with no new subscription.
Watch out: The episode was Zapier-sponsored; Zapier says it has a free-forever plan.
Goal: Let an agent call your n8n workflows like tools. The MCP Server Trigger turns an n8n workflow into an MCP server; clients connect by URL over SSE or streamable HTTP. Each workflow you expose becomes one tool.
For you: Cerebra or Alfred could trigger vetted n8n flows without being given n8n admin rights.
Watch out: n8n docs: behind a reverse proxy, turn off buffering and gzip for the MCP path or connections break.
Goal: Let an MCP client read, search and append to the vault. The Local REST API plugin opens an HTTPS API on 127.0.0.1:27124 with a bearer key, and now has a built-in MCP server. The separate mcp-obsidian server uses that plugin and offers tools such as search, get file, append and patch.
For you: Works only while Obsidian is open on that Mac; the agents on the VPS already use the GitHub copy of the vault instead.
Watch out: mcp-obsidian includes a delete_file tool; Chris's rule is archive, never delete, and the archive folder is off limits, so Sentinel should scope it first.
Goal: Use the same OpenRouter account in n8n as the agents do. n8n's OpenRouter Chat Model node plugs into its AI Agent nodes and lists only models your key can use. It has options for temperature, JSON output, timeouts and retries.
For you: Give n8n its own OpenRouter key with a spend limit, so its costs show apart from the agents'.
Plain-English meanings for the words the videos use, each with an everyday picture and where it shows up in Chris's own setup.
From your videos: Draws on 75 of the 105 videos, led by the beginner explainers (Jack Roberts, Sharbel A., Tina Huang, Simon Scrapes). News and repo-roundup videos add little here. Quantization and hallucination are not explained in any video. Plus 10 entries from the web to fill gaps (marked From the web): quantization, hallucination, temperature, fine-tuning vs prompting, system prompt, tool calling, structured outputs, knowledge cutoff, A2A protocol and guardrails.
An agent is an AI model that has been given tools, memory and a job. It can take steps on its own, not just answer one question. Tina Huang lists six parts: model, tools, memory, audio, guardrails and orchestration. Everyday picture: a new staff member with a desk, a phone and a to-do list, instead of a phone line you call with one question.
For you: Each of Chris's seven Hermes agents is one of these, each with its own lane (Alfred briefings, Cerebra research, and so on).
An agentic loop means the agent works, checks the result, and tries again until a goal is met. Hermes /goal keeps going until the goal is done, with a 20-turn budget by default per Jack Roberts. Simon Scrapes says Claude Code's /goal needs proof visible in chat. Everyday picture: a dishwasher cycle that keeps running until the dishes pass inspection, with a timer so it cannot run forever.
For you: Useful for Cyborg on a code fix with a clear test; Sharbel A. says to have Hermes write the goal prompt first.
Watch out: Loops spend tokens on every turn, so always keep a turn cap.
A channel is where you talk to the agent: Telegram, Discord, Slack, iMessage and more (Jack Roberts counts 22). The gateway is the Hermes service that links the agent to those channels and to the desktop app. Config changes need a gateway restart. Everyday picture: the switchboard that routes each phone line to the right person.
For you: Chris uses one Discord channel per agent, the same pattern Tina Huang uses.
Computer use lets an agent see the screen and move the mouse, like a person. Tina Huang calls this the agent's eyes. The Hermes HUD is a floating box that sees your screen, but only on the same machine. Everyday picture: a helper looking over your shoulder who can also take the mouse.
For you: Chris's agents live on a VPS, so screen features would not see his Mac without extra setup.
A cron job is a task that runs on a timetable, such as every morning at 09:00. A heartbeat is a regular wake-up where the agent checks for work. Sharbel A. says a good cron job is self-contained and returns a decision. Hermes cron jobs can now remember past runs. Everyday picture: an alarm clock with a note taped to it saying what to do when it rings.
For you: Brainiac's nightly filing and the n8n nightly note are both scheduled jobs; Alfred's briefings fit here too.
Watch out: Alex Finn says prune and pause old jobs; forgotten crons keep costing money.
The harness is the program wrapped around a model. It adds memory, tools, chat channels and schedules. Hermes, OpenClaw, Claude Code and Codex are harnesses. Paperclip is called a meta harness because it runs other harnesses as one team. Everyday picture: the model is the brain; the harness is the body, hands and diary.
For you: Hermes is the harness under all seven agents; Claude Haiku via OpenRouter is the brain inside it.
Watch out: Pav Rusovs says Anthropic rules bar using a Claude subscription through a third-party harness; that is his claim, so check current terms.
A task board with columns such as triage, to do, in progress, blocked and done. Hermes has one built in that hands cards to agents and sub-agents. Several presenters use it instead of a separate team tool. Everyday picture: sticky notes on a whiteboard that move left to right as work gets done.
For you: A research request could move from Cerebra to Brainiac to done, with Chris seeing every card.
One lead agent splits a job, hands parts to other agents and puts the answer together. Presenters often run the lead on a strong model and the workers on cheap ones. Dan Martell gives each worker one lane and a one-sentence definition of done. Everyday picture: a site foreman who reads the plans and hands jobs to the trades.
For you: Alfred could act as the single dispatcher; the native Hermes Kanban board is the place to try this first.
A Hermes profile is a separate home with its own keys, memory, schedule, skills and persona, so each profile is a different agent. Bot mode lets profiles chat in shared group rooms and @-mention each other. Alex Finn says run at least two so one can repair the other. Everyday picture: separate flats in one building: same address, own keys and own furniture.
For you: Chris's seven agents are seven profiles; a turn cap on group rooms keeps them from chattering.
Reverse prompting means telling the agent to ask you questions before it acts. Meta-prompting means asking the model to write the best prompt for the job first. Alex Finn runs a daily morning interview where the agent questions him and then proposes tasks. Everyday picture: a good tailor who measures you before cutting cloth.
For you: Alfred could run a short morning interview with Chris and turn the answers into Kanban cards.
A skill is a saved how-to file, usually markdown, that teaches an agent a repeatable job. It loads only when that job comes up. Sharbel A. keeps skills for procedures and memory for facts; Simon Scrapes says write down only what the agent gets wrong. Everyday picture: a recipe card you pull out only when cooking that dish.
For you: A skill could hold Chris's Coolify routine (Stop, untick Docker Cleanup, then Deploy). Every new skill goes to Sentinel first.
Watch out: Skills from strangers can carry hidden instructions; see Prompt injection.
SOUL.md is a plain text file that sets an agent's personality, role and rules. Dan Martell also uses an identity file (name and lane) and a user file (facts about you). Jack Roberts calls the soul file a character bible. Everyday picture: a job description and house rules pinned above the desk.
For you: Each of the seven agents has one; it is how Mantis stays walled off and Sentinel stays review-only.
Watch out: It loads every session, so a long soul file costs tokens on every turn.
A sub-agent is a short-lived helper the main agent spins up for one narrow task. Several can run side by side, and the main agent checks their work. The Hermes desktop app can now show, stop or steer running sub-agents. Everyday picture: a head chef sending three kitchen hands to chop, peel and wash at once, then tasting the result.
For you: Cerebra could send sub-agents to search several archives at once, then merge the findings.
Watch out: Each sub-agent is another model call, so parallel work multiplies spend.
The system prompt is the standing instruction text an app sends with every chat, kept apart from your own messages. It sets the role, tone and default behaviour for the whole conversation. Anthropic says even one sentence naming a role helps. Everyday picture: the job description pinned above a new worker's desk.
For you: Each Hermes agent's soul file plays this part; keeping it short leaves room for the actual work.
Checked 26 Oct 2026
A2A is an open standard for agents from different makers to find each other and share work. Each agent publishes an 'agent card', a small JSON file listing what it does and how to reach it. Google started it in April 2025 and donated it to the Linux Foundation, which now runs it as an open-source project. It complements MCP: MCP connects an agent to tools, A2A connects agents to agents. Everyday picture: business cards plus a shared work-order format between firms.
For you: Chris's agents talk through Discord inside one Hermes setup, so A2A is not needed today; it matters if he mixes agent products.
Watch out: IBM described it as early-stage in mid-2025; check what Hermes supports before planning on it.
Checked 26 Oct 2026
Agents can pass work to each other directly. Samin Yasar names the Agent Communication Protocol (ACP); Hermes bot mode uses @-mentions in shared rooms; Paperclip passes tasks through an org chart. Everyday picture: colleagues handing a file across the desk instead of through the boss.
For you: Chris's agents already post to each other's Discord channels; a turn cap keeps it bounded.
An API is a door that lets one program talk to another. An API key is the secret that proves it is you and bills your account. Treat it like a password. Everyday picture: a door and its key card; whoever holds the card gets in and you pay.
For you: Each agent has its own OpenRouter key; keys never go in Discord.
A hook is a small script that runs automatically at a set moment, such as before a prompt or when the agent says it is done. Unlike instruction files, hooks always run. Mason Fitzgerald uses one to inject memory; Sharbel A. uses one to trim command output. Everyday picture: a checkpoint gate that every car passes, whether the driver remembers or not.
For you: A hook could block any agent from reading the vault archive folder.
MCP is a standard plug that lets any agent use an outside app's tools in one common format. An MCP server offers the tools; the agent is the client. KodeKloud builds both in Python with FastMCP and tests them with MCP Inspector. Everyday picture: a USB port: any device with the right plug just works.
For you: The Hermes MCP catalogue is where an app like Canva or Notion would be switched on, after Sentinel's review.
Watch out: Each server adds token cost and trust risk; keep few and switch off unused ones.
corbin shows three ways to reach apps: MCP for chat-driven actions, an SDK for reusable scripts, and standalone connectors. He says a script is cheaper than MCP for repeat tasks because no model has to think each time. Everyday picture: phoning a shop each time versus setting up a standing order.
For you: Fits the Hermes cron no-agent mode: repeat jobs as scripts, models only for words.
Watch out: corbin's video is Zapier-sponsored.
OAuth lets you connect an app by signing in on the provider's own page, so the agent gets a limited pass, not your password. Composio and Granola MCP use it; Jack Roberts connects Grok with an X login. Everyday picture: a hotel key card that opens only your room, for a set stay.
For you: Connecting a calendar to Alfred this way means no password is stored in Hermes.
A plugin bundles several skills, tools or connectors into one install. Tina Huang shows Cowork plugins that combine skills and connectors. Hermes has a plugin catalog with badged, reviewed entries. Everyday picture: an app store bundle: one download, several features.
For you: Forge browses the Hermes plugin catalog; Sentinel reviews before install.
A slash command is a short typed instruction such as /model, /stop or /new. A CLI is a program you run by typing in a terminal, such as hermes doctor. Both control the agent without menus. Everyday picture: keyboard shortcuts for the whole agent.
For you: /new, /model and hermes doctor are the ones most useful for day-to-day upkeep.
Structured output makes a model answer in a fixed JSON shape you define, so another program can read it without errors. Anthropic enforces the shape during writing; n8n's OpenRouter node has a JSON response option. Everyday picture: a form with labelled boxes instead of a free-text letter.
For you: Useful when an agent hands data to n8n, such as a dashboard row or a filing record for Brainiac.
Watch out: Anthropic lists exceptions: refusals and answers cut off by the length limit may not match the shape.
Checked 26 Oct 2026
A tool is one action an agent can call: search the web, read a file, run a command. Harnesses let you switch tools on or off per agent; Wanderloots prunes them to stop tool loops. Everyday picture: one tool in a toolbox; a carpenter does not need the welding torch.
For you: Brainiac does not need browser automation; turning it off saves tokens.
Tool calling is how a model uses a tool. The app describes each tool with a name, a description and the inputs it takes. The model replies with a structured request to use one; the app runs it and sends the result back, and the model then answers. Everyday picture: a manager filling in a request slip that an assistant carries out.
For you: Every Hermes skill and MCP tool works this way; tool descriptions cost tokens on every call.
Watch out: Anthropic notes that tool definitions, tool calls and results are billed as input and output tokens, plus a small built-in tool system prompt.
Checked 26 Oct 2026
Instead of loading every tool's full description, the agent loads only a short index and fetches details when needed. Sharbel A. says this is the Claude Code default and roughly 85 percent cheaper. Everyday picture: a restaurant menu with dish names; you ask the waiter for details only on the one you want.
For you: The same idea as the skill retriever AI LABS shows for Hermes.
A webhook is a web address that another service calls when something happens, such as a new sign-up or payment. Sharbel A.'s watcher workflow reacts to them; Simon Scrapes says Claude routines can start from an API URL or a GitHub event. Everyday picture: a doorbell: you do not watch the door, it rings when someone arrives.
For you: An n8n webhook could alert Sentinel's Discord channel when a site event looks wrong.
Watch out: Webhook URLs are secrets; never post them in Discord.
Hermes keeps MEMORY.md (what the agent has learned) and USER.md (facts about you). Both load every session. Tina Huang calls these tier one of four memory tiers. Everyday picture: a sticky note on the fridge that everyone reads every morning.
For you: Chris's date format and 24-hour clock belong in USER.md so no agent has to ask.
Watch out: Because they load every time, keep them short.
Compaction squeezes a long chat into a summary (/compress in Hermes). /new starts a clean session. Jason Lee asks Claude for a handoff file at about 200 to 300 thousand tokens, then continues in a fresh chat. Everyday picture: writing a one-page shift handover so the next person need not read the whole logbook.
For you: One session per topic per agent keeps Haiku fast and cheap.
Context engineering is choosing what goes into the model's view for each job: the right files, rules and examples, and nothing extra. A Google AI engineer lists it among the four skills she hires for. Everyday picture: packing a suitcase for one trip, not taking the whole wardrobe.
For you: It is the thinking behind short soul files, scoped skills and one session per topic.
The context window is how much text a model can hold in mind at once. Each turn the whole conversation is sent again, so long chats get slow and costly. When it fills, old material is dropped or summarised. Everyday picture: a desk: once it is full, something has to go in a drawer.
For you: Long Discord threads with one agent slowly push early instructions out.
Dreaming is a background job where the agent reviews recent activity and promotes the important parts to long-term memory. OpenClaw has a dreaming toggle; Jack Roberts and Samin Yasar describe similar features in Hermes and Honcho. Everyday picture: tidying your desk at the end of the day and filing the keepers.
For you: Similar to Brainiac's nightly filing and the agents' daily diaries.
An embedding turns text into a list of numbers that captures meaning, so similar ideas sit close together. A vector database stores these and finds nearest matches by meaning. Mason Fitzgerald suggests text-embedding-3-small at about 5 dollars of credit. Everyday picture: a map where notes about similar things are pinned near each other.
For you: Qdrant is Chris's planned vector database for the vault; n8n's starter kit bundles it.
Watch out: Never index the archive folder.
Add-on services such as Honcho, Hindsight, Mem0 and Supermemory give agents deeper or shared memory. Sharbel A. says Hermes allows only one at a time. Everyday picture: a shared filing cabinet several staff can open.
For you: Hindsight overlaps the Qdrant plan, so pick one memory layer before building either.
Watch out: Shared memory services hold your data elsewhere; read what they send before connecting.
Eric Michaud splits his vault: agents may not edit the human side, and the machine side is theirs. Wanderloots adds an allow-list so agents touch only approved vaults. Everyday picture: a shared kitchen with one shelf labelled 'mine, do not touch'.
For you: Matches Chris's rule that the archive folder is off limits to every agent.
CLAUDE.md and AGENTS.md are always-loaded instruction files. Pav Rusovs keeps CLAUDE.md to about 40 lines as a map pointing to other files; his test took 30 seconds and 17k tokens with the map against 62 seconds and 28k without. Everyday picture: a building directory in the lobby, not the full blueprints.
For you: A short map file at the vault root could help every agent find things.
Watch out: Mason Fitzgerald calls CLAUDE.md a soft instruction; it is not enforced.
Andrej Karpathy's pattern: raw sources go in, and the model compiles them into linked, tidy markdown notes it keeps up to date. It has three jobs: ingest, query and lint, with an index file and an append-only log. Everyday picture: a librarian who reads every new book and updates the catalogue cards.
For you: Brainiac's vault filing already follows this shape.
RAG means: first look up the few most relevant passages, then give only those to the model. It lets a model answer from your own files without reading all of them. PrivateGPT and AnythingLLM do this locally. Everyday picture: a librarian who hands you the three right pages instead of the whole library.
For you: Cerebra answering from the vault by pulling the closest notes would be RAG.
Watch out: Mason Fitzgerald's speed and cost claims rest on his own 15-question test and a paid package.
Hermes stores past chats in a small database and can search them on demand. It is tier two of Tina Huang's memory tiers; Alex Finn calls it session recall. Everyday picture: searching your sent email instead of remembering every message.
For you: Lets an agent find what Chris decided weeks ago without stuffing it into every prompt.
A fallback chain lists backup models or accounts, so the agent keeps working if one fails or runs out. Hermes calls these fallback providers and credential pools. Everyday picture: a spare tyre in the boot.
For you: A cheap backup model per agent would stop a single provider outage from silencing the fleet.
Prompting means giving a model better instructions and examples each time. Fine-tuning means training a copy of the model on many examples so the behaviour is built in. OpenAI says prompting may be all you need, and fine-tuning helps mainly with large example sets, shorter prompts and cheaper small models. Everyday picture: a clear checklist for a temp worker vs sending them on a training course.
For you: Chris's agents run on prompts, soul files and skills; nothing in the setup needs fine-tuning.
Watch out: As of Oct 2026 OpenAI's page says its fine-tuning platform is winding down and closed to new users; existing users can still run training jobs for the coming months.
Checked 26 Oct 2026
A hallucination is when a model states something false or not supported by the material it was given, often in a confident tone. Anthropic's advice: allow the model to say 'I don't know', ask for direct quotes and citations, and limit it to the documents you supply. These steps reduce it but do not remove it. Everyday picture: a keen witness who fills gaps in memory with guesses.
For you: Cerebra's research and any family-history findings should carry a source link for each claim.
Watch out: Anthropic says to always check important facts yourself, especially for high-stakes decisions.
Checked 26 Oct 2026
A model only knows the world up to the date its training data stops. Anthropic lists a 'reliable knowledge cutoff' for each model; for Claude Haiku 5.5 it is June 2026. Anything newer must come from a tool such as web search or your own files. Everyday picture: a well-read friend who has been at sea since June.
For you: Cerebra's web search is how the fleet learns about anything after a model's cutoff.
Checked 26 Oct 2026
A cloud model runs on a company's servers and you pay per use. A local model runs on your own computer: private and free per use, but limited by your memory and chip. Tina Huang's rule: take 75 percent of your RAM and divide by 0.6 to get the billions of parameters that fit. Everyday picture: eating out versus cooking at home: home is cheaper per meal, but you buy the kitchen first.
For you: Chris's agents use cloud models today; a local model could take simple jobs if a Mac is sized for it.
Watch out: Local quality is lower, and hardware costs up front (Alex Finn prices a DGX Spark at about 4,800 USD).
Send the same question to several models, then merge or debate their answers. Hermes has /moa for this. Samin Yasar points to Karpathy's LLM Council, where models with personas argue to reach a better decision. Everyday picture: asking three doctors for an opinion before surgery.
For you: Worth it only for big decisions; it multiplies cost.
A gateway like OpenRouter gives you one account and one key for hundreds of models. You can switch models without new sign-ups, and set spend limits per key. AI Foundations reaches 312 models through one key. Everyday picture: one universal travel adaptor instead of a drawer full of plugs.
For you: Chris already uses OpenRouter for every agent.
Model routing means sending each job to the cheapest model that can do it well. Moe Lueker routes through OpenRouter to cheap models and keeps expensive ones for hard work. A Google engineer names it as a core skill. Everyday picture: you do not hire an electrician to change a light bulb.
For you: Haiku for filing and reminders; a stronger model only when Cerebra writes a long report.
The model is the AI that reads and writes text. Tina Huang sorts models into tiers: flagship, mid-tier, light and specialist (image, video, voice, code). Bigger tiers think better but cost more. Everyday picture: car engines: a race engine, a family engine and a scooter engine all move you, at very different running costs.
For you: All seven agents run on Claude Haiku, a light-tier model.
Watch out: Model names in auto-captions are often garbled; confirm a name on OpenRouter before switching.
An open model publishes its weights, so anyone can download and run it. Examples named include Qwen, DeepSeek, Kimi, GLM and Gemma. They can run locally or through hosted services. Everyday picture: a published recipe anyone can cook, versus a secret restaurant sauce.
For you: Possible cheap fallbacks on OpenRouter, after a test on one agent.
Quantization stores a model's numbers at lower precision, so the model needs less memory. Hugging Face says 8-bit roughly halves memory use with little quality loss; 4-bit shrinks it further. In GGUF files used by Ollama and LM Studio, a name like Q4_K means about 4 bits per weight. Everyday picture: saving a photo as a smaller JPEG; it fits on the phone, with slight blur.
For you: Only matters for local models, such as a private model for Mantis on the Mac; cloud models through OpenRouter are not affected.
Watch out: Smaller quants can make more mistakes; test on your own questions before trusting one.
Some models can think longer before answering. An effort setting lets you choose quick and cheap or slow and careful. The Hermes desktop app has an effort picker beside the model picker. Everyday picture: a quick glance versus a careful read before you sign.
For you: Keep effort low for routine filing and reminders; raise it only for hard one-off jobs.
Watch out: Changing effort mid-session resets the prompt cache, per Sharbel A.
Some models do one job: Whisper turns speech into text, Piper turns text into speech, and models like GPT Image 2 or FLUX make pictures. Tina Huang runs tiny ones even on a Raspberry Pi. Everyday picture: specialist tradespeople: a plumber, a painter, a sign writer.
For you: Chris already uses ElevenLabs for voice; Forge would own image and video jobs.
Some harnesses can sign in with a flat monthly plan (for example a ChatGPT plan via Codex login) instead of paying per token. Others use an API key billed per token. Jack Roberts mentions connecting the 20 dollar plan to Hermes. Everyday picture: a monthly transit pass versus paying for each ride.
For you: Chris pays per token through OpenRouter; a subscription route would be a separate decision.
Watch out: Pav Rusovs says Anthropic does not allow a Claude subscription through third-party harnesses; check current terms.
Temperature is a setting that controls how random a model's wording is. Low values give steadier, repeatable answers; high values give more varied ones. n8n's docs note that higher values also raise the risk of made-up facts. Ollama's default is 0.8. Everyday picture: a dimmer switch between 'stick to the script' and 'improvise'.
For you: Keep it low for filing and fact work (Brainiac, Cerebra) and only raise it for creative drafts (Forge).
Checked 26 Oct 2026
Uncensored local models have their safety refusals removed. Two videos show finding them by searching 'uncensored' in LM Studio or the Ollama library. Everyday picture: a car with the speed limiter taken out.
For you: Not a fit for the fleet; listed so the word is understood.
Watch out: The videos say little on safety or reliability; the dashboard notes treat this as a warning sign.
Everything loaded every turn (soul file, memory, skill descriptions, tool lists) is paid for every turn. Greg Isenberg's guest switches off unused built-in skills from about 150; AI LABS cuts skill context from about 11,000 to 2,300 tokens with a skill retriever. Jack Roberts' Context Doctor audits and quarantines rather than deletes. Everyday picture: carrying a full backpack on every errand.
For you: Context Doctor's quarantine approach suits Chris's archive-not-delete rule, after Sentinel's review.
A VPS is a rented server that runs all day. Metics Media prices a Hostinger KVM 1 at about 6 dollars a month; Lead Gen Jay puts Hetzner at about 5 to 6. Local hardware avoids API bills but costs up front. Everyday picture: renting a storage unit versus buying a garage.
For you: Chris's Hostinger VPS is the small line; the model bill is the big one.
Hermes can use one model for conversation and a cheaper auxiliary model for background tasks. Python Simplified picks a free Nemotron model on OpenRouter for the background work. Everyday picture: the senior partner meets clients; the junior does the photocopying.
For you: Background summaries and housekeeping could move to a cheaper model than Haiku.
Pay-per-token gateways bill only what you use. Flat plans cost the same each month: Metics Media prices Nous Portal at about 20 dollars a month. Alex Finn says Claude Opus costs him about 40 USD a day per token. Everyday picture: pay-as-you-go phone versus an unlimited plan.
For you: Chris is on pay-per-token Haiku via OpenRouter, with a budget that resets on 01 Nov 2026.
Re-sending the same opening text can be cheaper because providers cache it. Changing model, effort, MCP or plugin settings, or compacting, breaks the cache and costs more. Jay E shows Cache Tax, which warns once the cache has gone cold. Everyday picture: a pre-heated oven: open the door too often and you pay to heat it again.
For you: Avoid swapping an agent's model halfway through a session.
Some scheduled jobs need no AI at all. Hermes cron has a no-agent checkbox that runs a script and never calls the model. corbin makes the same point: a saved script beats an AI call for repeat tasks. Everyday picture: a timer switch on a lamp needs no electrician.
For you: Totals, backups and health checks can run as plain scripts.
OpenRouter lets you cap spend per key, so one runaway agent cannot drain the whole budget. Metics Media sets a weekly limit during Hermes setup. Everyday picture: a prepaid card for each teenager instead of sharing your credit card.
For you: One key and one limit per agent fits the seven-agent setup.
Run a cheap model by default and let it ask a stronger one only when stuck. Simon Scrapes describes /advisor in Claude Code, which does this without breaking the cache. Sharbel A. suggests Haiku sub-agents for cheaper side work. Everyday picture: a junior who phones the senior only for the hard questions.
For you: Haiku stays the default; a stronger model is called in for rare hard jobs.
A token is the unit models read and write in, a chunk of a word. Providers bill per token, both in and out. Because the whole conversation is re-sent each turn, long chats cost more each turn. Everyday picture: a taxi meter that ticks for every word, both ways.
For you: Every note Cerebra or Brainiac reads costs tokens on Chris's OpenRouter bill.
Many repos promise big savings. Caveman claims about 65 percent fewer output tokens; Nate Herk measured about 8.5 percent and no saving from RTK. Codebase-memory claims of 99 percent come from the project itself. Everyday picture: fuel-economy figures on the sticker versus on your own commute.
For you: Test any saver on one agent for a week before making it a fleet rule.
Meters show where tokens go. Sharbel A. points to /context, /usage and /cost in Claude Code; Nate Herk uses Codeburn to find idle MCP servers; The Next New Thing shows Model Deck for remaining subscription usage. Everyday picture: a smart electricity meter that shows which appliance is draining power.
For you: Per-agent spend could feed the Gauntlet dashboard.
An approval gate makes the agent stop and ask a person before it spends, sends or deletes. Sharbel A. sets Hermes to ask first; Alex Finn's Grok Bot asks before external actions; Pav Rusovs blocks send, share and delete in a settings file. Everyday picture: a second signature needed on any cheque.
For you: Matches Chris's rule: nothing installed without Sentinel's review and his yes.
A scheduled audit checks the agents for problems on a timetable. Tina Huang runs one twice a day; NetworkChuck watches Hermes crons, gateway and model login with Checkmk and sends alerts. Everyday picture: a smoke alarm plus a yearly inspection.
For you: A natural regular job for Sentinel, posting to her channel without secrets.
Jack Roberts backs up the whole Hermes agent to GitHub daily. Wanderloots uses Git save points for the vault so any agent change can be undone. Everyday picture: save points in a video game.
For you: The vault already syncs to GitHub; agent config, soul and memory files could too.
An eval runs a fixed set of test questions and scores the answers, so you know if a change helped. Tina Huang uses n8n Evaluations; Simon Scrapes uses a skill eval plugin that runs a job with and without a skill. Everyday picture: a driving test with the same route every time.
For you: Sentinel could check a score sheet before any prompt or skill change goes live.
Watch out: A model grading a model is a rough guide; spot-check by hand.
Guardrails are automatic checks on what goes into and comes out of an agent. In OpenAI's Agents SDK, an input guardrail screens the request and an output guardrail checks the final answer; if one trips, the SDK raises an error so the run stops. A cheap, fast model can do the screening before the costly one runs. Everyday picture: a bouncer at the door and a checker at the exit.
For you: Sentinel's reviews and the approval gates are Chris's human-side guardrails; an automatic check could sit in front of public-facing bots.
Checked 26 Oct 2026
Give each agent only the access its job needs. Nimrod uses read-only keys, read-only database users that cannot see personal data, and a local MCP proxy that exposes only allowed routes. Everyday picture: a cleaner gets a key to the front door, not the safe.
For you: Sentinel can check that each agent's keys are scoped to its lane.
Dashboards and admin pages should be reachable only through a private network, VPN or SSH tunnel, not the open internet. Tailscale links your own devices as if on one home network. Everyday picture: a staff-only door that needs a badge, instead of a shop window.
For you: Coolify and Hermes dashboards could sit behind Tailscale.
Prompt injection is hidden instructions planted in a web page, file or skill that try to hijack the agent. The Next New Thing reported a skill that steers users to a paid product from inside its own file. Jack Roberts shows Bouncer, which scans new skills and then allows, asks or quarantines them. Everyday picture: a forged note slipped into the in-tray saying 'pay this invoice'.
For you: This is why Sentinel reviews every skill before Chris says yes.
Watch out: Scanners miss things; a human still reads the skill file.
A chat bot can accidentally be opened to everyone. Python Simplified shows one environment setting that lets anyone talk to the bot, and fixes it by limiting admin commands to one user ID. Julian Goldie says the phone plugin is on by default in messaging apps. Everyday picture: leaving the office door propped open overnight.
For you: Sentinel can check every agent's channel settings for this.
A sandbox boxes the agent's code so a mistake cannot touch the rest of the machine. Wanderloots runs the Hermes terminal inside Docker; Dubibubi names Container Use for agent sandboxes. Everyday picture: a child's playpen: free to play, cannot reach the stove.
For you: Chris's agents already run in containers under Coolify.
Keys and passwords should never be pasted into chat, where they stay in history. Jack Roberts sets them as environment variables; Lead Gen Jay keeps them in Doppler, a central store. Everyday picture: keeping the spare key in a lockbox, not under the mat.
For you: Same as Chris's standing rule: no secrets in Discord.
Basic server safety: a non-root user, SSH keys instead of passwords, and a firewall that allows only SSH and web ports. Metics Media also turns on backups and changes default app logins. Everyday picture: locking the doors and windows on moving-in day.
For you: Applies to the Hostinger VPS under Coolify.
A turn cap limits how many back-and-forth messages agents may exchange before a human must step in. Wanderloots caps group rooms; Jack Roberts says /goal has a 20-turn budget. Everyday picture: a meeting with a hard stop time.
For you: Useful when Chris's agents talk to each other in Discord.
How the videos say to reach a Hermes agent from your phone, talk to it by voice, and in the newest release let it tap your phone's screen. Chris uses an iPhone, so each part flags what works on iPhone and what is Android only.
From your videos: About 20 of the 105 videos touch phone use. Only one, 'Hermes Agent Can Now CONTROL Your Phone!', covers true phone control, and its captions were Arabic so one name is low-confidence. Plus 12 entries from the web to fill gaps (marked From the web): Hermes's own platform docs, iMessage via BlueBubbles, ntfy and Pushover alerts, Shortcuts and Siri, Apple's iPhone Mirroring rules, ADB and mobile-mcp, and Tailscale on phones.
Apple's Shortcuts app on iPhone and iPad has a 'Get Contents of URL' action that makes a web request and can send JSON, a form or a file. That is how a Shortcut can hand a message to an agent's webhook or notification topic.
For you: iPhone and iPad only. Pairs with ntfy or an n8n webhook.
Checked 26 Oct 2026
Siri runs any shortcut when you say its name, on iPhone, iPad, HomePod, Apple Watch or Mac. Combine that with the ntfy channel: a shortcut that posts your words to the agent's ntfy topic, and the reply comes back as a push.
For you: Hands-free for driving, next to the ElevenLabs voice work; iPhone only.
Watch out: If the topic name or token is inside the shortcut, anyone you share the shortcut with can command the agent. Never share it.
Checked 26 Oct 2026
Many presenters (Alex Finn, Tina Huang, Teku AI, Pav Rusovs) use Tailscale, a private encrypted network, so a phone or laptop can reach the VPS and home machines without opening anything to the public internet.
For you: A safer way for your iPhone to reach the Hostinger VPS or a home Mac than exposing a port.
Tailscale's app works on iOS 15 or later (iPhone and iPad) and Android 8.0 or later. On iPhone you accept a VPN configuration and push notifications, which warn you when you need to log in again.
For you: Lets the iPhone reach a private Hermes dashboard or self-hosted ntfy without opening them to the internet.
Watch out: Allow its push notifications, or an expired key can silently cut the phone off.
Checked 26 Oct 2026
Julian Goldie lists four ways to control Hermes from a phone: 1) Telegram, the simplest; 2) Discord, best for teams; 3) a self-hosted browser dashboard kept behind a VPN or SSH tunnel; 4) a community Android app where you secure your own server.
For you: Options 1 and 2 are iPhone-friendly; option 4 is Android only.
Watch out: The donts list says never put a dashboard on the open internet; reach it over Tailscale or a tunnel only.
ADB (Android Debug Bridge) is Google's official tool for controlling an Android device from a computer. Android 11 and later can pair over Wi-Fi with a QR or pairing code; the phone and computer must be on the same network.
For you: Android only; not relevant to Chris's iPhone unless he adds a spare Android phone.
Watch out: Ticking 'always allow on this network' makes that network trusted for debugging. Use the Wireless debugging quick tile to switch it off when done.
Checked 26 Oct 2026
For an Android phone, Phone Harness connects over ADB (Android Debug Bridge, the standard USB/Wi-Fi developer link). There is also a cloud phone option from Phone Harness cloud.
Watch out: Android only: the 'back' command works here but not on iPhone. You do not use Android, so this is for reference.
Malware families such as ADB.Miner, Ares, IPStorm, Fbot and Trinity have scanned the internet for open ADB connections and installed malware on the devices they found.
For you: A Sentinel check: ADB must never be reachable from the internet; Tailscale or USB only.
Watch out: Revoke ADB debugging authorisations on the phone when you stop using an agent with it.
Checked 26 Oct 2026
For an iPhone, Phone Harness connects through iPhone Mirroring on a Mac. The app that runs Hermes needs Accessibility and Screen Recording permissions turned on.
For you: This is the iPhone path. It needs one of your Macs on and mirroring.
Watch out: iPhone has no 'back' command: Hermes taps the app's on-screen back arrow instead. iPhone control needs a Mac in the loop.
Apple's iPhone Mirroring lets a Mac show and control the iPhone. It needs macOS Sequoia 15 or later on a Mac with Apple silicon or a T2 chip, iOS 18 or later with a passcode, both signed in to the same Apple Account with two-factor authentication, Bluetooth and Wi-Fi on, and the iPhone locked and near the Mac. The Mac must not be sharing its internet connection or using AirPlay or Sidecar.
For you: This is what the Phone Harness iPhone route relies on, so it only works from a Mac beside the phone, never from the VPS alone.
Watch out: Whatever can control the Mac can then control the iPhone too, including banking apps.
Checked 26 Oct 2026
Mobile Next's mobile-mcp server lets an agent drive iOS simulators, Android emulators, and real iOS or Android devices. It needs Node.js 20 or later, Xcode command line tools for iOS and Android Platform Tools (adb) for Android, on the computer it runs on. It is built mainly for app testing and automation.
For you: Useful for Cyborg/Vic's iOS/Swift work on a simulator; a real iPhone needs developer tooling on a Mac.
Watch out: Its install line runs the latest npm package each time; Sentinel should pin and review a version first.
Checked 26 Oct 2026
Julian Goldie shows Hermes v0.22.0 (13 Oct 2026) adding phone control through a community plugin called Phone Harness. It gives Hermes one tool, phone_exec: Hermes writes a short script and the plugin runs it on a phone. The author is named only with low confidence and no repo was shown.
For you: Exactly the kind of thing Sentinel must review and you must approve before it goes near your setup.
Watch out: The plugin is on by default in messaging apps, so anyone who can message your bot could drive your phone. Turn it off per platform with the hermes tools command, or skip the plugin. This is on your 'don't' list.
Julian Goldie's safety rules: every phone script asks before it runs, you allow it once or always, and it never types your passwords, PINs or two-factor codes. Keep approvals on.
For you: Lines up with your standing rule that nothing risky runs without your yes, and no secrets ever get handled loosely.
iMessage is a built-in Hermes channel, named by Tina Huang, Alex Finn and others as a quick way to reach the agent from an Apple device. The videos show it as a channel to switch on, but give no step-by-step for it.
For you: Works on your iPhone because iMessage is Apple's own app.
Watch out: Coverage is thin: no video walks through iMessage setup end to end.
Hermes can use iMessage through BlueBubbles, a free open-source server that runs on a Mac signed into Messages. Messages flow iMessage to Messages.app to BlueBubbles to Hermes, and back. The docs recommend DM pairing: a new sender gets a pairing code that you must approve.
For you: Chris's Hermes runs on the VPS, so this needs a Mac running BlueBubbles that the VPS can reach (for example over Tailscale).
Watch out: By default Hermes answers every authorised iMessage DM or group message; turn on require_mention for groups. iPhone only; Android has no iMessage.
Checked 26 Oct 2026
Lead Gen Jay names Photon (photon.dot.codes, as spoken) as a way to give an agent free iMessage access without needing a Mac mini running. It is mentioned only, with no setup shown.
For you: Of interest because your Hermes lives on a Hostinger VPS, not a Mac.
Watch out: Named once, from audio; confirm the site and have Sentinel review before use.
Hermes lives inside chat apps you already use, so your phone becomes the remote. Across the videos the presenters name Telegram, Discord, WhatsApp, iMessage, Slack, Signal, email and Matrix as channels. Telegram is the one most recommend for a phone.
For you: You already run a Discord channel per agent, so your agents are reachable on your phone through Discord today.
Metics Media and Sharbel A. show the quick path: open the Channels page, pick Telegram, scan the QR code with your phone, and it makes a private bot only you can talk to. Set it as your home channel so the agent sends updates there.
Watch out: Jack Roberts and Python Simplified warn a public bot can be opened to everyone; keep admin commands tied to your own ID.
Sharbel A.'s tip: put the bot in a Telegram group and turn on topics, giving you rooms by job (general, email, alerts, content) while keeping one agent and one memory. He says this one change makes it far more usable on a phone.
For you: This matches your 'newest entry on top, one file' habit: one agent, tidy rooms, nothing lost.
The Hermes docs say one background process, the gateway, connects every chat platform you set up: Telegram, Discord, Slack, WhatsApp, Signal, SMS, email, iMessage (BlueBubbles or Photon), ntfy and more. It also runs cron jobs and delivers voice messages. Voice works on Telegram, Discord and Slack (and some others), but not on Signal or the standard WhatsApp bridge; plain SMS has no voice, images or files.
For you: Chris already uses Discord, which the docs mark as supporting voice, images, files, threads and reactions. Works the same on iPhone and Android.
Checked 26 Oct 2026
If you only want Hermes to push alerts (cron summaries, job done) and never take commands back, the docs say to set NTFY_TOPIC and NTFY_PUBLISH_TOPIC to the same value and leave NTFY_ALLOWED_USERS out. A cron job can then send with: hermes send ntfy:alerts-channel "Done!"
For you: A safe first step for Alfred: one-way reminders to the iPhone, no inbound control.
Watch out: No threads or attachments on ntfy; long replies stay in one notification.
Checked 26 Oct 2026
ntfy is a simple push-notification service. Hermes supports it as a channel: subscribe to a topic in the ntfy phone app, send a message to that topic, and the agent's reply arrives as a push notification. It works with the free ntfy.sh server or a self-hosted one.
For you: Works on iPhone and Android. Self-hosting on Coolify would keep it private.
Watch out: The Hermes docs say anyone who knows the topic name can talk to the agent. Use a self-hosted topic with access control, a reserved topic with a token, or at least a long unguessable name, and send nothing sensitive otherwise. ntfy caps each message at 4,096 characters; Hermes cuts longer replies.
Checked 26 Oct 2026
Pushover sends notifications to iPhone, iPad, Android, desktop and Apple Watch through a small API. It costs $4.99 USD once per platform after a 30-day free trial; an iPhone and iPad share one licence.
For you: An option if Chris wants a polished app rather than ntfy; Hermes's docs list ntfy, not Pushover, as a built-in channel.
Watch out: Its API token is a secret; keep it in the server's environment, never in Discord.
Checked 26 Oct 2026
One video shows ChatGPT 'remote access' letting you drive your laptop from your phone, and another shows Telegram voice notes driving a home Mac mini. This is phone-to-computer control, the reverse of Phone Harness.
Watch out: Screens 5 is shown only for iPad-to-Mac remote desktop; these are separate tools, not Hermes features.
The videos name a few phone-side apps: Dispatch for chatting to Claude Cowork from a phone, T3 Code's mobile app (Nimrod runs 100+ projects with it), and Grok Bot, which offers an iPhone action-button call and iPhone app control. These are mentions, not setup guides.
Watch out: Grok Bot and T3 Code are other products, not Hermes; treat as alternatives, not part of your stack.
Tina Huang shows Google AI Edge Gallery, an iPhone app to download and run Gemma 4 on the phone, and Draw Things for on-phone image generation. These run on the device, not through Hermes.
For you: iPhone apps, so they fit your device; a way to try a private model with no server.
Alex Finn shows ChatGPT voice mode wired to Hermes for back-and-forth speech. It needs an OpenAI API key.
Watch out: Alex Finn says it costs about 5 cents a minute, so it is not free.
Nimrod records a voice note with the iPhone Action button and it goes to Hermes, where Whisper transcribes it; Alex Finn uses Telegram voice notes the same way. One video shows Telegram voice notes driving a home Mac mini.
For you: The iPhone Action-button shortcut is iPhone-friendly and fits how you like to work away from the screen.
Metics Media and Alex Finn show the Hermes desktop app's heads-up display (Cmd+Shift+H) with voice and a 'Hey, Hermes' wake word, so you can speak to the agent hands-free.
Watch out: Teku AI says the HUD only works on the same machine as the agent, so it is a desktop feature, not a phone one.
Short recipes for one job each, taken from the videos. Pick a job, follow 3 to 8 steps, then check the result.
From your videos: 27 recipes from 18 of the 105 videos. Hermes is well covered; n8n, Tailscale and Discord-specific steps are thin, because the videos rarely show the clicks. Plus 11 entries from the web (marked From the web) filling the Discord, n8n and Tailscale gaps from official docs.
Simon Scrapes says skills should hold only what Claude gets wrong, and should be tested with and without.
For you: Short skills keep Haiku costs down across all seven agents.
Sharbel A. shows four built-in views and a check on connected tools.
Watch out: The caption drops the panel's command name; the notes say it is almost certainly /mcp.
Sharbel A. says almost all usage is the agent re-reading history, not what you type. These habits cut it.
For you: Forge and Cyborg/Vic do the coding work; Chris's agents already run on Haiku.
Watch out: He says compacting is the most expensive message of a session and wipes the cache.
Hermes's channel and user settings need Discord IDs, not names. Developer Mode adds a Copy ID option to right-click menus.
For you: Needed for DISCORD_ALLOWED_USERS and DISCORD_ALLOWED_CHANNELS in each agent's settings.
Hermes has settings that limit where a bot answers. Set an allowed channel list, and the bot replies only there; DMs are handled separately.
For you: Matches Chris's channel-per-agent layout; Mantis's private channel is a good first candidate for a tight allowlist.
Watch out: Leave DISCORD_ALLOW_MENTION_EVERYONE and DISCORD_ALLOW_MENTION_ROLES off (false is the default), or a model reply with @everyone pings the whole server.
Checked 26 Oct 2026
Hermes's Discord page gives the full path: a Discord app, a bot user, two switches turned on, a token, and an invite link. Repeat it once per agent so each has its own name and token.
For you: One bot per agent (Alfred, Brainiac, Cerebra and so on) keeps each agent's channel and token separate.
Watch out: Without Message Content Intent Discord refuses the connection and the bot stays offline. Never post the token in Discord.
Checked 26 Oct 2026
A private channel is hidden from everyone except the roles or members you add. You need admin rights on the server.
For you: Good for Mantis (walled off): only Chris and the Mantis bot see the channel.
Hermes can add one outside memory service on top of its own files. Tina Huang uses Honcho; Sharbel A. names Mem0, Hindsight and Supermemory as other options (spelled as captioned).
For you: Decide between a provider like this and the planned Qdrant index before building either.
Watch out: Only one external provider can be active at a time, per Sharbel A.
A profile is a separate Hermes with its own keys, memory, cron jobs and personality. Alex Finn recommends at least two so one can fix the other.
For you: Chris already runs seven; this is how a spare 'repair' profile could be added.
Watch out: Sharbel A. lists building profiles before workflows as a mistake: know each profile's job first.
Tina Huang explains the second memory tier: Hermes stores past chats in a database and searches them only when asked.
Python Simplified turns a blank Hermes into 'Amber', a customer service agent, by replacing SOUL.md and pointing it at a folder of business documents.
For you: One soul file per agent keeps each of the seven in its lane. The video deletes the old file; keep a copy in the archive instead.
Watch out: The presenter stresses telling the agent to use the knowledge folder; uploading files alone does not make it read them.
Tina Huang has Hermes save a research result into her vault with the Obsidian skill, then pulls a stored document back to plan the next step.
For you: This is Brainiac's lane. The vault archive folder stays off limits to every agent.
Metics Media shows a correction in one chat carrying into a brand-new session, because Hermes writes it to memory.
Watch out: Sharbel A.: memory is for stable facts and preferences, not task progress or to-do lists.
A skill is a written procedure Hermes loads when the task matches. Tina Huang builds a business-idea scorer this way, and Sharbel A. calls the correct-and-update loop Hermes's superpower.
For you: Each agent can grow its own small skills; Sentinel reviews any skill that came from outside.
Watch out: Sharbel A. separates the two: memory holds facts, skills hold procedures.
Use a strong model for hard work and a cheap one for background jobs. Python Simplified sets a free model for 'auxiliary' tasks; Sharbel A. says you can just tell Hermes which model to use for what.
For you: Chris runs Claude Haiku via OpenRouter; confirm any model name on OpenRouter's own list before switching an agent.
Watch out: Sharbel A.: a weak model on hard multi-step work can cost more in retries than a strong one.
/goal keeps Hermes working until the goal is met. Sharbel A. first asks Hermes to design the goal prompt, then answers its questions.
For you: Good for Cyborg/Vic's code work; keep goals read-only until checked.
Watch out: Jack Roberts says /goal runs with a turn budget; keep a cap so it cannot loop forever.
Sub-agents are helpers that work on parts of a job at the same time. Metics Media plans a trip with four of them; Sharbel A. says each needs a narrow job and a check.
For you: Cerebra's research jobs split well this way.
Watch out: Sharbel A.: trusting sub-agent output without checking is one of the seven common mistakes.
Sharbel A. puts Hermes in a Telegram group with topics turned on, so one agent has separate rooms that do not bury each other.
For you: Chris already does this with one Discord channel per agent.
Metics Media shows the slash command that changes the model from Telegram or any surface with no buttons.
Watch out: Sharbel A. explains for Claude Code that switching model mid-session throws away the cache and re-bills the whole chat; pick the model at the start where you can.
Wanderloots prunes tool toggles per bot to stop tool loops. Alex Finn uses hermes tools to switch browser and computer use, and Julian Goldie uses it to switch off the phone plugin per platform.
For you: Sentinel can keep the list of what each agent is allowed; Mantis should have the fewest tools.
Watch out: Julian Goldie says the phone plugin is on by default in messaging apps, so anyone who can message the bot could use it.
Metics Media's troubleshooting order: read the logs, restart the gateway, then restart the whole app.
For you: On Coolify, do the app-level restart as Stop (untick Docker Cleanup) then Deploy, not Restart.
NetworkChuck monitors his Hermes agent with Checkmk and sends an alert to Slack when its gateway goes down, then a recovery message.
For you: Methods named in the video include Slack and Mattermost; Discord was not shown, so check before planning a Sentinel-channel alert. Never put secrets in the alert text.
Python Simplified opens a customer bot to everyone, then limits admin commands to his own Telegram ID.
For you: Chris's bots are private, so skip the allow-all step; the admin-only block still helps.
Watch out: The allow-all variable opens the bot to anyone on Telegram.
Metics Media sets a weekly cap on the OpenRouter key Hermes uses, so a runaway job cannot surprise you.
For you: One key per agent, each with its own limit, matches the per-key cost view planned for Chris's dashboard.
The Hermes quickstart gives a fixed order for troubleshooting, from health check to gateway.
For you: Complements 'Fix a bot that stopped replying'.
Checked 26 Oct 2026
In two setup videos the presenter describes an outcome once and Hermes writes the skill, sets the schedule and only pings when something is new.
For you: Fits Cerebra (web research). Sentinel should read any dependency it installs.
Watch out: Metics Media notes Hermes may install a dependency on its own; it asks permission for consequential steps, so read the prompt before you approve.
A cron job is a task Hermes runs on a timer. Sharbel A. says each run starts in a fresh session, so the prompt must stand on its own. A good job says what to check, where, what counts, what to ignore and where to send the result.
For you: Alfred owns briefings; deliver each job to that agent's own Discord channel, not a shared one.
Watch out: Sharbel A. lists 'making every idea a cron job' as a top mistake: a job with no clear decision becomes notification spam.
Alex Finn says pruning and pausing old cron jobs makes Hermes faster and cheaper. Greg Isenberg's guest shows the cron panel in the desktop app.
For you: Chris's rule: archive, never delete. Save the prompt to the vault archive before removing a job.
Python Simplified runs a Python script on a schedule and only uses the model when judgement is needed. Ticking no-agent runs the script with no AI credits at all.
For you: Chris is on Eastern Time; check the server clock with date before setting any hour. Good for Sentinel or Alfred jobs that only count things.
Watch out: The folder name matters: the video says Hermes only finds the script if the folder is called 'scripts'.
Both Sharbel A. videos open with this job: Hermes asks your number one priority each morning, then helps with it. Its last line keeps it from spending or sending on its own.
For you: A natural Alfred job, posting to Alfred's Discord channel.
Watch out: Keep the check-in short; the presenter says the goal is leverage, not more work.
n8n can run a second 'error workflow' whenever a workflow fails. One error workflow can watch many workflows.
For you: Pairs with the Hermes gateway alert how-to so both n8n and agents report failures to one place.
Checked 26 Oct 2026
For one-way messages (reports, alerts) a Discord webhook is enough: no bot user, no login. It can only post, not listen.
For you: Fits an n8n alert channel; agents that must reply still need their own bot.
Watch out: Anyone with the webhook URL can post into that channel, so treat it like a secret.
The Schedule Trigger starts a workflow every so many minutes, hours or days, or on a cron pattern (a 5-part time rule). It runs only when the workflow is saved and published.
For you: Whitby is in the same time zone as New York, but set America/Toronto explicitly so it is clear.
Watch out: If a cron rule uses a variable, changing it later does nothing until you unpublish and republish.
Checked 26 Oct 2026
A Webhook node gives the workflow its own web address. A test address works while you listen in the editor; the real one works only after you publish.
For you: A way for a Hermes agent or a phone shortcut to kick off an n8n job; pair it with a schedule for a daily run plus on-demand runs.
Watch out: Payloads over 16 MB are refused unless N8N_PAYLOAD_SIZE_MAX is raised.
Checked 26 Oct 2026
Tailscale Serve puts a local web page on an HTTPS address that only your own Tailscale devices can open. Funnel is the public version; do not use it for dashboards.
For you: Fits Chris's rule that dashboards are reached over Tailscale or a tunnel only (see the 'Don't put dashboards on the open internet' warning).
Watch out: One port is either Serve (private) or Funnel (public): the last command wins. Funnel is enabled on devices by default, so turn it off if unused.
Checked 26 Oct 2026
Device keys expire after 180 days by default (for new tailnets), and an expired server disappears from the tailnet. Tailscale suggests turning expiry off for trusted servers.
Watch out: If a key does expire, do not run tailscale up --force-reauth over a Tailscale-only SSH session; you can cut yourself off. Keep the Hostinger browser terminal as a way in.
Checked 26 Oct 2026
Wanderloots clips a page into the raw folder, then asks his agent to turn it into linked wiki notes and run maintenance.
For you: This is Brainiac's filing job; Git is the undo button if an agent gets it wrong.
Dan Martell builds two documents: a 'brain system' about how you think, and a voice style guide.
For you: Forge (editing) could use the style guide.
Watch out: The prompts are in his video; the notes do not quote them word for word.
Use cases where AI makes investing easier for someone in Ontario: research, tracking, learning, scam checks, tax records and paper-only experiments. Information only, not financial advice.
From your videos: 18 of the 105 videos feed this tab, most with a passing mention, and only one (Samin Yasar's course) shows a trading build, on paper. Plus 21 entries from the web to fill gaps (marked From the web): Canadian regulators, accounts and brokers, CRA tax, filings research, back-testing, crypto data and scam warnings.
The Next New Thing reports an Anthropic Financial Services repo with ten finance agents, such as KYC (know-your-customer checks), plus knowledge-work plugins for finance and other teams. Tina Huang shows a Cowork finance plugin with skills like audits and close management.
For you: Aimed at finance teams; no video uses them for personal investing.
Watch out: Named only; no demo of results is shown. Built for finance teams and largely US workflows; nothing shown is tailored to Canadian accounts.
Tina Huang says she runs a team of local agents researching stocks she might buy, next to a software team for her company. The agent logs what it researched so later chats about those stocks have context.
For you: Cyborg/Vic with Cerebra doing the web research.
Goal: understand a company from its own documents, not from social media. SEDAR+ is the CSA's free filing system for Canadian public companies: annual reports, interim financial statements and press releases. EDGAR is the US version; the SEC's data.sec.gov APIs need no API key and return filing lists and financial figures as data.
For you: Cerebra does the reading; Brainiac files the summary in Obsidian.
Watch out: SEDAR+ company search defaults to the last six months, so widen the date range to find annual reports. data.sec.gov is US filers only.
Checked 26 Oct 2026
Goal: let an agent look up coin prices and exchange data without scraping websites. CoinGecko's API covers over 18,000 coins and 1,500 exchanges, has a free Demo plan with limited endpoints and rate limits, and lists MCP servers and agent setup guides.
For you: Read-only data for Cyborg/Vic; no wallet or exchange keys needed.
Watch out: The Demo plan's exact limits are not on the docs home page; check before scheduling frequent jobs. Data only, not a signal to buy.
Checked 26 Oct 2026
Matt Wolfe and others report new models built to return choices, scores or probabilities instead of text: OpenAI's Decisions API, Perplexity's open decision models, Microsoft Decision One and Nimble V4, and Jev.
Watch out: News items only; none of the videos tests them on markets.
Sharbel A. lists a deal research desk and a real-estate deal scout among 50 Hermes services. The agent monitors listings, compares areas and prices, and builds memos and decision briefs. He says it works when the research is tied to a real financial decision.
For you: Cerebra-style research; useful for a home or property decision.
Watch out: A service idea; the video shows no working build.
Last 30 Days searches Reddit, X, YouTube, Hacker News, Polymarket (a prediction market) and the web for what people said about a topic in the past month.
For you: Cerebra could run it weekly on one or two tickers Chris follows.
Watch out: It reports chatter, not facts; Sentinel should review the skill before install.
Goal: get the main points of a company's quarter in five lines. Canadian companies post interim financial statements and press releases on SEDAR+; US companies file theirs on EDGAR, whose data APIs update in real time as filings come in (financial figures usually within a minute).
For you: Cerebra writes it; Alfred can drop it into the next morning brief.
Watch out: AI summaries can miss or invent numbers; check every figure you act on against the filing.
Checked 26 Oct 2026
The Next New Thing notes a new version of TimesFM, Google's model for forecasting numbers over time.
Watch out: Named only; no finance example is shown.
Tina Huang keeps a 'portfolio' project folder with a dashboard in Claude Cowork. She asks Sonnet which holding is her best and worst performer, and it answers from her files.
For you: Cyborg/Vic could read an exported holdings file, with no bank logins.
Watch out: She stresses this is not financial advice, and admits her China AI holdings are 'all underwater'.
Goal: hear about big moves or news on a few names without watching screens. Questrade's MCP server gives an agent read-only quotes, historical prices and account data; ntfy or Pushover can push the alert to your iPhone.
For you: Cyborg/Vic checks; Alfred sends the alert. Keep the Questrade link read-only.
Watch out: Pushover is $4.99 USD one time per platform after a 30-day trial. Too many alerts teach you to ignore them.
Checked 26 Oct 2026
Tina Huang's scheduled tasks run every day at 8:00 p.m.: an investment deep dive, a portfolio briefing and a macro briefing. They feed an investments dashboard she built. In another video she describes an investment dashboard with custom news and curated suggestions for her holdings.
For you: Cyborg/Vic writes it, Alfred delivers it in the daily brief.
Tina Huang's digest pulls calendars, Gmail, investments, Slack and Notion and writes it to Apple Notes. Her local-agent video lists 'priority stock portfolio' among morning brief items.
For you: Alfred's briefing, with a short market line from Cyborg/Vic.
Goal: see all accounts in one place and get a nudge when your mix drifts from your plan. The agent reads an exported holdings file or a read-only broker link, compares it to your target mix, and reminds you; you make any trade yourself.
For you: Cyborg/Vic compares; Alfred reminds; Brainiac files the monthly note.
Watch out: Each account has its own tax rules and contribution room; an agent should never move money between them.
Checked 26 Oct 2026
Questrade offers an MCP server so Claude, Claude Code, ChatGPT, Cursor, VS Code and Codex CLI can read balances, positions, orders, quotes and option chains. Trading needs a separate permission: the agent can only draft an order preview, and you must approve it in the Questrade app before it goes to market.
For you: The most Canadian-friendly option for Cyborg/Vic: start read-only and leave trade permission off.
Watch out: Drafting orders needs Questrade app 2.0.0 or later with Push Approval turned on. Order and activity history covers only the last 90 days. A confirmed cancellation cannot be undone. Sentinel should review before connecting.
Checked 26 Oct 2026
Goal: understand what you own and what it costs. Canadian ETFs must give buyers an 'ETF Facts' document that shows costs (including the MER, the yearly fee as a percent), past returns, holdings and risks. An agent can read it and explain it plainly.
For you: A good first job for Cerebra: reading an official document, no account access.
Watch out: Use the official ETF Facts, not a blog's summary; fees change.
Checked 26 Oct 2026
Tina Huang lists investing as the first AI skill for 2026. Her view: index funds are already heavy in AI, so decide how much AI exposure you want given your job. She is heavily exposed through her career, so she hedges. She shares a prompt in her description to help think it through.
For you: A question for Chris to think about, not an agent job.
Watch out: She says plainly this is not financial advice and not a finance channel.
Goal: understand terms like MER, ACB (adjusted cost base) or rebalancing at your own pace. An OSC study found people followed AI investment suggestions about as closely as human ones, and warned AI advice can be biased, irrelevant or inaccurate.
For you: Any agent can do this; keep it separate from account access.
Watch out: Learning is fine; treat any 'buy this' answer as unverified, because the same study shows people trust AI suggestions readily.
Checked 26 Oct 2026
Alex Finn says his agent-run site has made $1,100, all from two sponsorships, and talks of several such businesses one day replacing a job. Dubibubi's agent-run Etsy shop made $4.55 in 10 days.
Watch out: Neither is investing, and both show how little proven income sits behind the hype.
In October 2024 the US SEC settled charges against Rimar Capital, which raised nearly $4 million (USD) from 45 investors while falsely saying it used an AI platform to trade. Shown here as a US example of the pattern Canadian regulators also warn about.
Watch out: 'Uses AI' in a pitch proves nothing; ask for registration, not technology claims.
Checked 26 Oct 2026
Goal: catch a scam in two minutes. The OSC warns that scammers sell 'AI trading bots' that may not use AI at all, let you withdraw small amounts first to build trust, use deepfake celebrity videos, and push urgency. Its first tip: check whether the person or firm is registered.
For you: Sentinel's kind of job: review a claim, never act on it.
Watch out: CIRO says it will never ask for your banking details or ID copies; anyone who does while claiming to be a regulator is a fraudster.
Checked 26 Oct 2026
Sharbel A. lists spending money as one of three things a bot must never do without him. He sets gates in three layers: don't give the tool, use Hermes's approval setting, and write the rule in SOUL.md as a softer backup.
For you: Already Chris's rule; worth writing into Cyborg/Vic's soul file too.
Watch out: A SOUL.md rule is not a hard wall; removing the tool is.
Paul J Lipsky reports that Claude Duet asks approval before payments or deletions, and will never trade stocks or move money between accounts, even with approval.
Watch out: A vendor drawing a hard line on money moves is a useful benchmark for your own agents.
In an OSC experiment, people put 22% more money into AI-enhanced scams than into ordinary ones. Learning about scams first cut that by 10%, and a browser plug-in that flagged scams cut it by 31%.
For you: Background for Sentinel's scam checks.
Watch out: A simulated test, but it shows warnings and tools do help.
Checked 26 Oct 2026
Goal: keep the records the CRA expects. The CRA says selling, trading, gifting, converting or spending crypto can be taxable; profits are business income if you trade as a business, otherwise capital gains. Keep records of every buy and sale and how you worked out fair market value.
For you: Cyborg/Vic builds the ledger; Brainiac files it yearly. No exchange keys with trade rights.
Watch out: The CRA notes its crypto guide is being updated; check the current version. Unreported gains bring tax, penalties and interest.
Checked 26 Oct 2026
If you trade often in a TFSA, the CRA may treat it as running a business and tax the profits as business income. It looks at trade frequency, short holding periods, time spent and systematic strategies; prior years can be reassessed.
Watch out: Relevant to any trading bot: an automated strategy in a TFSA looks systematic by design.
Checked 26 Oct 2026
Alpaca's official MCP server lets Claude and other agents check accounts, pull market data and place orders. It defaults to paper trading; live trading needs ALPACA_PAPER_TRADE=false and live keys. Alpaca says orders go straight to its trading API, so review AI-suggested orders.
Watch out: Not for Canadians yet: on 14 Aug 2026 Alpaca staff said that, because of Canadian regulatory issues, it does not offer market data to Canadian residents; it is working toward individual accounts for Canadians and targets regulatory approval around November 2026. US market data only.
Checked 26 Oct 2026
LEAN is QuantConnect's open-source trading engine (Apache-2.0) in Python and C#. Its CLI, installed with pip install lean, runs back-tests, optimisation and research notebooks locally in Docker.
For you: A Cyborg/Vic study project; never run its live mode without Sentinel and Chris.
Watch out: A good back-test does not predict future returns; the copy-trading entry's 57.8% is exactly such a back-test.
Checked 26 Oct 2026
In the same course, Samin connects a strategy site heard as Quiver Quant through its API as an MCP. It lists disclosed trades by US politicians and others. He has Hermes copy one politician's trades into a $10,000 paper portfolio, check the site daily during trading hours, and post a dashboard every Friday.
For you: A study project for Cyborg/Vic on paper only.
Watch out: Trades are mirrored with a 1 to 3 day disclosure lag, and the 57.8% three-month return he shows is the site's own back-test, not his result. Canada: this idea uses US data (US politicians' disclosed trades). Quiver's site lists a $25 a month or $300 a year premium plan without naming the currency (assume USD). The nearest Canadian source is SEDI, the public record of insider trades at Canadian companies, which covers company insiders (not politicians) and posts filings within about 5 minutes.
Samin Yasar's three-hour Hermes course has a trading-bot build. He connects Hermes to a broker platform heard as Co-invest (co-invest.ai) through an MCP with OAuth, turns on paper trading, and talks to it in a Telegram trading topic. Every reply comes with an HTML dashboard of the positions. He says the platform covers stocks, crypto, forex and prediction bets.
For you: Cyborg/Vic covers markets; a paper account and read-only access would match Chris's rule that nothing is paid without his yes.
Watch out: It is a paper-trading demo, and at one point his bot 'just bought a LEAP of Uber for some reason' that he could not recall asking for. Canada: the video does not say whether this platform accepts Canadian residents, so check its sign-up page first. Keep it on paper.
IBKR clients get a paper trading account (found in Client Portal settings) to test strategies and its APIs before going live. For Canadians there is a firm limit: residents cannot send orders through their own or third-party apps via the API for products on Canadian exchanges; IBKR cites CIRO dealer rules.
Watch out: Paper fills can differ from real ones. An agent placing orders on TSX-listed products through the API is not allowed for Canadian residents.
Checked 26 Oct 2026
Questrade offers a practice account for its Questrade Global platform with $100,000 CAD in virtual cash. Sign-up asks for name, email and phone.
For you: A Canadian place to practise by hand before any agent is involved.
Watch out: The page says 30 days in one place and 90 in another; confirm the length. The page does not say the MCP server works with it.
Checked 26 Oct 2026
Questrade: official MCP server, read-only by default, orders need your approval in its app; a Questrade Global practice account exists. Interactive Brokers Canada: full API and paper account, but no API orders on Canadian-exchange products for residents. Wealthsimple: its own site shows no developer API or paper account; 'Wealthsimple MCP' servers online are third-party.
For you: If Cyborg/Vic ever gets a broker link, Questrade's read-only MCP is the narrowest one.
Watch out: Never give an unofficial tool your Wealthsimple or bank login.
Checked 26 Oct 2026
TFSA: no deduction, but growth and withdrawals are tax-free. RRSP: contributions cut taxable income now; withdrawals are taxed later. FHSA: contributions are deductible, growth is untaxed, and qualifying withdrawals for a first home are tax-free.
Watch out: Limits change each year; check your current room with the CRA, not with an AI.
Checked 26 Oct 2026
The Canadian Investor Protection Fund returns your property if a member dealer goes insolvent, up to $1 million for general accounts (cash, margin, TFSA, FHSA), $1 million for retirement accounts (RRSP, RRIF, LIF), and $1 million for RESPs. Amounts in CAD. It does not cover market losses, bad advice, fraud, or crypto missing at insolvency. Claims must be filed within 180 days of the insolvency.
Watch out: Losses from an AI bot or a scam are not covered; nor is a platform that is not a CIPF member.
Checked 26 Oct 2026
The OSC regulates securities in Ontario and is one of the provincial members of the CSA, which runs SEDAR+ and the National Registrant Search. CIRO oversees investment dealers and keeps AdvisorReport. Before trusting any firm or app that touches your money, check it is registered.
Watch out: A US-registered broker is not automatically allowed to serve Ontario residents (see the Alpaca entry).
Checked 26 Oct 2026
Mini-courses built from the longest teaching videos Chris saved: the multi-hour courses and full guides first, then the best 'in X minutes' explainers. Each topic is a short, ordered set of lessons you can follow one after another.
From your videos: Draws on roughly 45 of the 105 videos, led by the long ones: a 3-hour Hermes course, two 1-hour courses, and several 20-to-31 minute guides. Hermes, Claude Code, local AI and memory are richly covered; AI video is lighter. Plus 30 lessons from official docs and free courses to fill gaps (marked From the web), including two new web-only topics: Prompting well, and Swift and iOS with AI. AI video now has 5 web lessons (Remotion, HyperFrames, ElevenLabs, HeyGen Academy, LTX-2).
The Hermes docs split learning into three tiers: Beginner (about 1 hour: install, quickstart, CLI, configuration), Intermediate (2 to 3 hours: sessions, messaging, tools, skills, memory, cron) and Advanced (4 to 6 hours: architecture, custom tools and skills). There is also a 'team of specialist Bots' route through Desktop, Profiles, Bot Mode and Cron.
For you: The 'team of specialist Bots' route is the closest match to your seven-agent crew.
Sharbel A. and Metics Media both run the same zero-to-working path: host Hermes (your own machine, or a one-click Hostinger/Nous Portal VPS), connect a model, then connect Telegram so it reaches your phone. Alex Finn's desktop walkthrough covers the same ground.
For you: Your Hermes already lives on a Hostinger VPS under Coolify, so this is the ground you built on.
The docs quickstart says to get plain chat working first, then add the next layer (gateway, tools, skills, a Docker sandbox). Hermes needs a model with at least 64K tokens of context. Secrets go in ~/.hermes/.env and other settings in ~/.hermes/config.yaml.
The courses teach the mental model: a SOUL.md persona file, built-in memory files (USER.md, MEMORY.md) for facts, and skills for repeatable procedures kept separate from facts. Sharbel A. and Jack Roberts lay this out; the 3-hour course builds self-made skills on top.
For you: Each of your seven agents keeps its own lane this way.
Built-in memory is two small files in ~/.hermes/memories/: MEMORY.md (the agent's notes, 2,200 characters) and USER.md (your profile, 1,375 characters). They load as a frozen snapshot at session start, so a change shows only in the next session. Memory does not auto-compact: when full, the agent must tidy it before writing more. Writes are scanned for injection and leaked credentials.
For you: Write approval suits Mantis and any agent that touches private notes.
Watch out: External memory plugins (Honcho, Mem0 and others) run alongside built-in memory; have Sentinel review any before install.
A skill is a folder with a SKILL.md file (YAML front matter, then instructions), stored in ~/.hermes/skills/. Every hub install passes a built-in scanner for data theft, prompt injection and destructive commands, and gets a trust level: builtin, official, trusted or community. A 'dangerous' verdict can never be forced through.
For you: This is the gate Sentinel should sit behind before you say yes to any skill.
Watch out: The scanner helps but does not replace a human read; the videos still say to read skills from unknown authors line by line.
The most-named Hermes feature. Sharbel A.'s classic is a morning cron that asks your number-one priority and starts work; Metics Media shows the agent writing its own watch-and-report skill and schedule.
Watch out: Alex Finn says prune and pause old cron jobs; they cost speed and money.
Run more than one agent. Profiles are separate agents with their own keys, memory, cron and persona; sub-agents are narrow helpers the main agent spins up and checks. Alex Finn recommends at least two profiles so one can repair the other.
For you: This is the backbone of your seven-agent team.
Wanderloots teaches bot mode: specialist bots in group rooms with a turn cap and human approval, plus the native Kanban board for assigning, approving and replaying multi-agent work. Metics Media and Sharbel A. point to the Kanban board as the native way to run a team.
Watch out: Wanderloots prunes each bot's tool toggles to stop tool loops; the donts list says try the native board before any extra orchestration layer.
In the docs, a Bot is just a Hermes profile with its own model, memory, skills, credentials and chat history under ~/.hermes/profiles/<name>/. Bot routines are ordinary cron jobs named [bot:<name>] <routine>. Bots can share group chats of 2 to 6 and message each other with the message_agent tool. The docs page does not mention a Kanban board.
For you: Each of your seven agents maps to one profile, so their memories stay apart.
Anthropic's Claude Academy has short free Claude Code tutorials: What are skills? (3 min), Creating your first skill, Configuration and multi-file skills, Sharing skills, Troubleshooting skills (4 min each) and Using Claude Code Remote Control. A roundup from April 2026 also lists longer free Academy courses: Claude Code 101, Claude Code in Action, Introduction to Subagents and Introduction to Agent Skills.
Watch out: The four longer course pages were not opened directly; check them on Anthropic's site before you start.
Simon Scrapes's discipline: try a job with no skill first, then write down only what Claude gets wrong. He describes an Anthropic skill-eval plugin that runs the same job with and without a skill so you can see if it helps. A skill that changes nothing is just extra tokens.
In Claude Code a skill is a folder with SKILL.md: front matter (name, description) then the steps. Personal skills live in ~/.claude/skills/<name>/, project skills in .claude/skills/<name>/. The docs say keep SKILL.md under about 500 lines and move long reference material into other files in the same folder. Claude loads a skill when the description matches, or you call it with /skill-name.
Dubibubi and Nate Herk tour the reference sets: Anthropic's own skills repo, Superpowers (brainstorm, plan, test-driven build, review, debug), gstack's role skills, Matt Pocock's pack (Grill Me, Teach Me), and Planning with Files, which keeps the plan on disk.
Watch out: Nate Herk says star counts and token-saving claims are marketing until you test them on your own work.
Mason Fitzgerald's point: CLAUDE.md is a soft instruction nothing forces Claude to obey; hooks are the hard version, small scripts that fire on events. Simon Scrapes uses stop hooks to run checks the moment Claude claims it is done; Jay E tours mods like Blast Radius that pause risky deletes.
The official docs say CLAUDE.md instructions are advisory, while hooks are scripts that always run at set points (for example after every file edit, or before a tool runs). Use a hook for anything that must happen every time with no exceptions. Claude can write the hook for you.
Pav Rusovs keeps CLAUDE.md to about 40 lines, a map pointing to signpost files, not a manual; he claims a map cut one task from about 28k to 17k tokens. Nimrod uses a two-line global rule: look in the wiki first, write notes to the inbox.
For you: Fits your lean, plain-file approach to the vault.
Run /init to get a starter CLAUDE.md, then cut it down. For each line ask: 'Would removing this cause Claude to make mistakes?' If not, cut it. A bloated file makes Claude ignore real rules. Run /context to check it loaded. Keep sometimes-needed knowledge in skills instead.
Jason Lee has Claude Design (the /design canvas) mock up app screens from a reference design system, then builds the app, landing page and a launch video in one run, using a handoff file to restart in a fresh chat near the token limit.
For you: Matches your wish to see a draft design before anything is built.
Anthropic's best-practice guide: use plan mode (Shift+Tab) to explore and plan before code, then give Claude a check it can run (tests, a build, a screenshot) so it can fix its own mistakes. If you have corrected it twice on the same thing, run /clear and start again with a better prompt. Before calling work done, have a subagent review the diff with fresh eyes.
Tina Huang's 24-minute guide steps from an AI second brain, to an AI database where the agent writes notes, to Karpathy's LLM Wiki run by a Hermes agent. She names the Smart Connections and Copilot community plugins and paid Obsidian Sync for sharing one vault across machines and agents.
For you: Your vault 'The Cone of Silence' is exactly this kind of agent memory.
Obsidian properties are small labelled fields (text, list, number, checkbox, date, tags) stored as YAML at the top of a note between --- lines. Once a property name has a type, it keeps that type across the whole vault, so agents can sort and filter notes reliably.
For you: Brainiac can file notes with the same properties every time, which keeps The Cone of Silence searchable.
Wanderloots and Tina Huang build Karpathy's pattern: a raw-sources folder, an LLM-owned wiki folder, and a schema file, with an index.md catalogue and an append-only log.md. A skill ingests sources into cited wiki notes.
For you: The one-file, newest-first log matches your recurring-log rule exactly.
Bases is a core Obsidian plugin that shows notes as a database view: tables, lists, cards, kanban boards or maps, built from note properties. Everything stays in your Markdown files. A base is saved as a .base file or embedded in a note.
For you: A base over agent diary notes could be a simple at-a-glance view, with no extra plugin to install.
Eric Michaud splits the vault so agents may not edit the human side; an agents file at the root of every folder states the rules, and agents log what they did to the daily note. Wanderloots adds grounded citations.
For you: Lines up with your archive-not-delete rule and agents being walled off from the archive root.
Pav Rusovs and Wanderloots schedule a lint job that reports broken links and bad front matter rather than trusting the agent to keep the vault tidy.
Watch out: The donts list: don't build a wiki and never lint it.
Tina Huang's 'every way to run open models' goes easiest to hardest: Ollama on your own machine, browser playgrounds (Arena, Groq, Hugging Face Spaces), borrowed GPUs (Google Colab), managed APIs (Together, Fireworks), then your own VPS or rented GPUs.
Ollama runs models on your own machine and serves them at http://localhost:11434. The quickstart's small local model is about 7.2 GB, with a baseline of 8 GB of free graphics memory (or unified memory on a Mac). With less, Ollama can use system RAM but answers are slower. Larger context windows need more memory.
Her 'every size' guide gives a rule of thumb (usable RAM is about 75% of RAM, divide by 0.6 for billions of parameters) and shows Qwen 3 as a go-to local LLM, down to tiny models on a phone or microcontroller.
Watch out: A small local model is a 'grunt worker', not a genius; keep a strong model for the real thinking.
Hermes can run local models (Settings > Providers detects your hardware and suggests a fit, such as Gemma 4). Tina Huang shows OpenClaw pointed at Ollama, and the self-hosted starter kit bundling n8n, Ollama, Qdrant and Postgres.
For you: A private-lane option for Mantis without a cloud model.
Ollama offers an OpenAI-style endpoint (/v1/chat/completions) and an Anthropic-style one (/v1/messages), so many agents can point at it unchanged. The docs warn each layer covers only part of the original API. It can also start Claude Code against Ollama models with 'ollama launch claude'.
Watch out: Keep any Ollama cloud API key on the server, never in browser code or a Git repo.
n8n Learn lists four courses: Essentials: Your First Workflows, Integrations: APIs and Connected Workflows, In Practice: AI, Testing and Best Practices, and an n8n Quickstart. Registration is free; the page does not state course prices.
The no-code n8n build names the components: a model, tools, memory, audio, guardrails and orchestration, assembled in a visual workflow. A meta-prompt in ChatGPT writes the agent's prompt.
The n8n docs group AI into agents, chains, tools, memory and vector databases, built on LangChain ideas. You can connect providers such as OpenAI, Anthropic and Google, and mix several models in one workflow.
AI Foundations builds an n8n research agent: a query refiner, a style guide, a research manager and an operations manager that splits tasks, storing sections and citations in Airtable (or Notion / Google Sheets) and outputting a PDF. OpenRouter gives one key for many models.
For you: Close to your own n8n research workflow and nightly-note setup.
Corbin's guide shows three ways to give an agent thousands of apps through Zapier (MCP, SDK, standalone connectors), demoed by creating a calendar event. Add n8n evaluation nodes to test the agent against set cases.
Watch out: The donts list: never let an agent send, spend or delete without a human step.
An n8n AI Agent node can use MCP servers as tools. The node panel has a registry you can connect in one click: pick the server, sign in, choose which tools the agent may use. For other servers, the MCP Client Tool node connects by URL and credentials.
Watch out: Only tick the tools the agent needs; every extra tool is extra reach.
n8n has two kinds of evaluation: light checks during building (run the workflow against known test cases) and metric-based scoring for workflows already in use. The section page does not give setup steps; they are on its sub-pages.
For you: A few known test questions would catch it early if your research workflow quietly gets worse.
OWASP's standard list of the ten biggest risks in apps built on language models: LLM01 Prompt Injection, LLM02 Sensitive Information Disclosure, LLM03 Supply Chain, LLM04 Data and Model Poisoning, LLM05 Improper Output Handling, LLM06 Excessive Agency, LLM07 System Prompt Leakage, LLM08 Vector and Embedding Weaknesses, LLM09 Misinformation, LLM10 Unbounded Consumption (runaway cost or resource use).
For you: Sentinel's reviews can use these ten as a checklist.
Published 9 Dec 2025, this list is about agents that act: ASI01 Agent Goal Hijack, ASI02 Tool Misuse, ASI03 Identity and Privilege Abuse, ASI04 Agentic Supply Chain (bad MCP servers, connectors, prompts), ASI05 Unexpected Code Execution, ASI06 Memory and Context Poisoning, ASI07 Insecure Inter-Agent Communication, ASI08 Cascading Failures, ASI09 Human-Agent Trust Exploitation, ASI10 Rogue Agents.
For you: ASI06 and ASI07 map straight onto shared vault memory and agents messaging each other in Discord.
Metics Media's VPS tutorials: a non-root sudo user, SSH keys (ed25519), a firewall (UFW) allowing only SSH, 80 and 443, and backups on from day one. Dokploy or Coolify-style panels sit on top.
For you: Your VPS runs Coolify; the same hardening rules apply.
The Hermes security docs: risky shell commands go through approvals (mode smart by default, or manual, or off), a hard blocklist always refuses things like disk wipes, and timed-out approvals are denied. Chat gateways deny everyone unless you set an allowlist. Writes to SSH keys and secret stores are blocked.
Watch out: The docs say the deny rules reduce accidents but do not sandbox a hostile agent; approvals off or YOLO mode removes most prompts.
Nimrod gives each agent a scoped, read-only key, read-only database users that cannot see personal-data tables, and a small MCP proxy exposing only allowed routes. Nate Herk makes data stores read-only by default.
For you: Matches your Sentinel-reviews-everything and no-secrets rules.
Jack Roberts: never paste API keys into chat, use environment variables, keep tokens in a central store like Doppler, and back up each agent to GitHub daily. Sharbel A. and Pav Rusovs gate anything that spends, sends or deletes.
For you: Your no-secrets-in-Discord rule and GitHub-synced vault already follow this.
Tina Huang avoids registry skills unless the author is trusted; Metics Media reviews Skills Hub items first. A skill is instructions the agent obeys, so it can hide a push; Jack Roberts demos Bouncer and Nvidia's Skill Inspector for prompt-injection checks.
Watch out: The donts list: don't read star counts as a safety rating, and don't install an unread skill.
Sharbel A.'s three tiers: a top model for hard thinking, a mid model for everyday work, a cheap or free model for grunt work (sorting, tagging, summarising). A model being free does not make it smart. Simon Scrapes's /advisor lets a cheap model ask a stronger one only when stuck.
Python Simplified shows a no-agent checkbox on cron jobs so script-only jobs run without calling a model, plus a cheap auxiliary model for background work alongside the main one.
For you: Useful before your 01 Nov budget reset.
Sharbel A. explains Claude Code re-sends the whole conversation each turn; switching model, effort or MCP servers mid-session resets the cache and costs far more. Start fresh with /new or /clear when context fills.
Watch out: The donts list: one session per topic, avoid mid-session toggles.
For Anthropic models on OpenRouter you mark cache points yourself with cache_control (up to four). The cache lasts 5 minutes by default, or 1 hour with ttl set to 1h. OpenRouter keeps sending the same model to the same provider for 10 minutes so the cache stays warm.
For you: Your agents run Claude Haiku through OpenRouter, so this is where the cache savings would show.
Jack Roberts and Metics Media put a per-key spend limit on OpenRouter so one runaway agent cannot drain the fleet's budget, and measure any token-saving claim on one agent for a week before it becomes a fleet rule.
For you: Each of your agents already has its own OpenRouter key.
OpenRouter keys can carry a credit limit, and limit_reset can set it to reset daily (midnight UTC), weekly or monthly. Keys report daily, weekly and monthly usage. There are also guardrails you can assign to keys, though the page does not describe their rules.
For you: A daily limit per agent key stops one runaway agent from eating the month's budget.
openrouter/auto picks a model per prompt based on what OpenRouter users spent on for that kind of task over the last 7 days. You can set a cost_tier (low to max) and limit choices with allowed_models. You pay the normal rate of the model it picks.
Watch out: It follows what others spend on, not what is cheapest, so set cost_tier and allowed_models if you use it.
Tristen O'Brien shows Claude Motion turning a prompt into an editable, code-based animation you refine with tweaks and comments, pairing it with Higgsfield and the free Remotion as a code-driven alternative.
Watch out: The video says Motion was beta and limited to Team/Enterprise plans; confirm availability before relying on it.
Jason Lee uses a /brag-style skill to turn a finished project into a short launch video as coded motion graphics, pasting a reference video so Claude mimics its animation style.
Watch out: The repo for the /brag skill was not named on screen; treat as a pattern, not a confirmed install.
Remotion's own guide shows how to make a video just by prompting a coding agent such as Claude Code, Codex or OpenCode. You need Node.js and a coding agent; most agents need a paid plan. There is also a 5-minute beginner video on the page.
Watch out: Remotion uses its own source-available licence, not a standard open-source one (per the HyperFrames comparison). Check it before business use.
HeyGen's open-source HyperFrames makes video from plain HTML files, with no build step. It ships agent skills, a catalogue of ready blocks (transitions, captions, charts) and a browser studio. Its README compares it with Remotion: same headless-Chrome-plus-FFmpeg method, but HTML instead of React, and an Apache 2.0 licence with no per-render fees.
ElevenLabs' quickstart lets a coding agent do the setup. You paste one line into the agent; a browser opens once to sign in and authorize. The key is saved in the project's env file and never shown in the chat. If the agent runs on a remote machine, you create the key yourself and paste it into the env file.
For you: Matches Chris's no-secrets-in-chat rule; Hermes voice already uses ElevenLabs.
HeyGen's free Academy is a short course on making avatar videos: platform overview, avatars, voices, translation into other languages, the Video Agent and AI Studio. It also has business modules and playbooks for marketers and trainers.
Watch out: The '10x faster' claim is HeyGen's own marketing.
Lightricks' LTX-2 makes video and matching sound in one model, with open weights. You can try it in the LTX Studio playground, run it in ComfyUI with the LTXVideo nodes, or use the Python code. Its tips: width and height must divide by 32.
For you: The local code needs an NVIDIA GPU (CUDA), so on Chris's Mac use the online playground.
Watch out: It uses Lightricks' own open-weights licence, not a standard one. The card says it is not for factual content and audio without speech can be weaker.
Anthropic's guide: treat Claude like a brilliant new employee with no context. Test: show the prompt to a colleague; if they would be confused, Claude will be too. Give the reason behind a rule, for example 'this will be read aloud', instead of a bare 'NEVER do X'.
For you: Useful when you write SOUL files and skills for the agents.
Examples steer format and tone better than description. Use 3 to 5 that are relevant and varied (include edge cases), and wrap each in <example> tags, inside <examples>, so Claude does not mix them up with instructions. You can ask Claude to check your examples or write more.
Put instructions, context and input in their own tags, such as <instructions>, <context> and <input>. For long documents, put the documents first and the question last, and ask Claude to quote the relevant parts before answering.
Tell Claude what to do, not what to avoid: 'write in flowing paragraphs' works better than 'don't use markdown'. 'Can you suggest changes?' may get only suggestions; 'Change this function' gets edits.
Since 15 Sep 2025 you can sign in to Claude inside Xcode 26 for a coding assistant that sees your project, plus tools to write docs, explain code, make SwiftUI previews and edit inline. It works on plans that include Claude Code (Pro, Max, and premium Team or Enterprise seats), and usage limits are shared with your other Claude use.
For you: Fits Cyborg/Vic's iOS and Swift work, without a separate tool to install.
Apple's Foundation Models framework gives Swift apps the Apple Intelligence model on the device, plus Private Cloud Compute for bigger jobs. The @Generable macro makes the model return your own Swift types, and the Tool protocol lets it call your app's functions. Users need a device that supports Apple Intelligence.
Watch out: Only Apple Intelligence devices can run it, so plan a fallback for older iPhones.
A private home base for you and your agents that only you can reach: self-hosted apps, a monitoring page, agent dashboards and a wiki, joined by a private network. This tab gathers what the videos show and how they keep it private.
From your videos: 26 of the 105 videos feed this tab. Only bashbunni's video (XkAsmjwE8h4) uses the word "intranet"; the rest are the closest things: self-hosted apps, NetworkChuck's homelab monitoring, private agent dashboards, Tailscale-only access and Obsidian as a wiki. No video builds a single intranet home page that links everything; that part is a gap. Plus 14 entries from the web to fill gaps (marked From the web): five home page dashboards (Homepage, Homarr, Glance, Dashy, Heimdall), privacy gates (Tailscale Serve, Cloudflare Access, authentik, Authelia), Uptime Kuma and Beszel, web wiki options, and the Hermes dashboard and hermes-webui login rules. Coolify's one-click list has Homepage, Homarr, Dashy, Glance, Uptime Kuma, authentik, Wiki.js, Docmost and Beszel.
Julian Goldie lists four ways to control Hermes remotely, including a self-hosted browser dashboard reached only through a VPN or SSH tunnel. Pav Rusovs links laptop, phone and server with Tailscale so the server has no public address. Alex Finn uses Tailscale so Hermes can reach all his devices.
His Hermie HQ mission control is hosted on Vercel but gated behind a Google login, so only he can open it. It shows check-ins, Kanban cards from his morning workflow and his agents' work.
hermes dashboard runs on 127.0.0.1:9119 with no login by default. If you bind it to any other address, it refuses to start until a login is set up: username and password (for trusted networks or VPN only), Nous Portal OAuth (recommended for internet-facing), or your own OIDC such as authentik. The docs recommend keeping it on loopback and reaching it by SSH tunnel or Tailscale.
Watch out: The Kanban docs say plugin routes are not authenticated, so --host 0.0.0.0 exposes the board.
Checked 26 Oct 2026
On a Hostinger VPS the Hermes dashboard can be put behind HTTPS with a Traefik option. It shows sessions, models, cron, skills, keys, channels and logs, and the system page can stop, start or restart the gateway.
Watch out: HTTPS encrypts the page but does not hide it; add a private network or a login gate as well.
nesquena's Hermes WebUI binds to 127.0.0.1 by default and its password is off by default. You turn it on with HERMES_WEBUI_PASSWORD or in Settings; passkeys and OIDC login are also supported. The README recommends an SSH tunnel or Tailscale for remote use and a password before exposing the port.
For you: Chris runs Hermes web UIs on Coolify; worth checking each one has a password or passkey set.
Watch out: The README notes that reverse-proxy basic auth can leave the installed phone app (PWA) on a blank screen.
Instead of public chat apps, some presenters host the place they talk to agents: Nimrod uses Element on his own Matrix server, and Samin Yasar deploys Buzz on a VPS as a shared workspace for people and agents.
A CTO keeps an old PC running his agents around the clock and reaches it over SSH and Tailscale with MagicDNS, or T3 Connect for a few devices. He chats with his Hermes agent through a self-hosted Element and Matrix server, and his agents get scoped read-only keys.
Local models with Ollama, a chat workspace in AnythingLLM or Open WebUI (which can link to n8n and MCP connectors), and remote use from an iPad through Screens 5. He loaded his about-me and writing-rules files into AnythingLLM.
For you: Chris already runs Open WebUI at chat.agent86.cloud; this shows the same idea kept at home.
Watch out: His voice dictation app stopped working offline.
A $0 home lab that replaces cloud subscriptions, all reached over Tailscale. She edits video on her home machine remotely, syncs her Org-mode notes to her phone through a self-hosted web app, and streams her phone camera into OBS. She rejected Parsec because it has no LAN-only mode and no Tailscale integration.
For you: Tailscale Serve is the pattern for putting the Gauntlet or Hermes web UIs on Chris's devices without a public address.
Watch out: Only Arabic auto-captions were available for this video, so some app names were checked by sound.
PrivateGPT runs on your own machine at localhost:8001 and answers questions over your PDFs with RAG. n8n's self-hosted AI starter kit bundles n8n, Ollama, Qdrant and Postgres for a private agent stack.
A day-one self-hosting stack: Jellyfin or Plex for media, Navidrome for music, Kavita for books, Audiobookshelf for audiobooks, Immich for photos (an iCloud replacement) and Paperless-ngx for documents. Portainer manages the Docker containers, Nginx Proxy Manager gives each app a domain name, and Tailscale is the private way in.
For you: Coolify already does Portainer's job for Chris; Immich and Paperless-ngx are the apps closest to his iCloud and vault habits.
Watch out: Some names came from garbled captions (Paperless-ngx is a best guess).
A 2FA and single sign-on server that works as an add-on to your reverse proxy. It has second factors such as one-time codes, security keys and passkeys, and Duo push, plus lockout after too many tries. Its docs list Traefik, Caddy, NGINX and others as supported proxies.
For you: Coolify uses Traefik, which Authelia supports, but Authelia is not in Coolify's one-click list.
Watch out: Lighter idea than authentik but needs proxy settings per app.
A self-hosted identity provider (one login for many apps). It speaks OAuth2/OIDC, SAML, LDAP and RADIUS. Core is MIT, with a separate paid enterprise part. Installs with Docker Compose for small setups. About 26.1k stars.
For you: Homarr, Hermes's dashboard and hermes-webui can all use OIDC, so one authentik login could cover them. It is in Coolify's one-click list.
Watch out: One more server to secure and back up; worth it only once several apps need logins.
Checked 26 Oct 2026
Access puts a login wall in front of a self-hosted app on a public name. Every app is deny by default; only people matching an Allow policy get in, through a login provider you pick (for example Google or GitHub) and optional MFA. The app is connected through a Cloudflare Tunnel.
For you: Only for a page that must be reached without Tailscale; donts.json says dashboards stay off the open internet.
Watch out: Order matters: a tunnel without the Access app first is open to anyone with the link.
Metics Media's VPS basics: a non-root sudo user, SSH keys only, and a UFW firewall that allows only SSH, 80 and 443. Change app defaults on day one; Umami ships with the login admin/umami.
Cloudflare Tunnel can expose a local app to the internet, which Tina Huang uses for demos. Public dashboards in the videos sit behind a login, as with Sharbel A.'s Google gate.
Watch out: A tunnel makes the app reachable by anyone with the link unless you add a login.
One setting can open a Hermes bot to everyone. Python Simplified restricts admin commands to the owner's own ID. Julian Goldie warns the phone-control plugin is on by default in messaging apps. Nimrod uses scoped read-only keys.
Every home-lab and agent video that covers access uses a private network rather than open ports. bashbunni chose tools that work LAN-only or over Tailscale; Pav Rusovs keeps the server off the public internet.
Serve shares a local web app with the devices on your own tailnet (your private Tailscale network) only. Tailscale gives it an HTTPS certificate for its tailnet name. Funnel is the different command that makes it public on the internet.
Watch out: Use Serve, not Funnel, for intranet pages; Funnel is public.
Checked 26 Oct 2026
Privacy also means knowing what is running. Checkmk alerts when a service stops, and Tina Huang schedules agent security audits that report on a timer.
A notification rule sends problems to Slack (Mattermost and email are other options); when he stopped an agent's gateway, the alert arrived in about a minute, and a recovery message followed. A second rule sends gateway failures to Paperclip as a ticket, and an IT agent restarted the gateway on its own.
For you: Chris's equivalent would be an alert into an agent's Discord channel, with no secrets in the message.
A light server monitor: a hub web app plus a small agent on each machine. It keeps history of CPU, memory, disk and network per Docker container, and alerts on CPU, memory, disk, temperature and more. OAuth/OIDC login, backups to disk or S3. MIT, about 26.6k stars.
For you: A lighter choice than Checkmk for one Hostinger VPS; it is in Coolify's one-click list.
Watch out: Installing the agent on the VPS needs Sentinel's review first.
Checked 26 Oct 2026
NetworkChuck installs the free Checkmk community edition on a VM in under five minutes and monitors Linux servers, network gear, Docker containers, Proxmox VMs, websites and a Windows PC. Data is collected every minute and kept for up to four years, with default good and bad thresholds. Every change must be activated, a bit like committing code.
For you: Could watch the Hostinger VPS, Coolify containers and the agent86.cloud sites from one page; it is a new install, so Sentinel reviews it first.
Watch out: The free edition needs manual work (plugins, piggyback folders) that the paid editions automate; the video was sponsored by Checkmk.
Once you know how Checkmk works, you can give an agent an automation account (a machine user with an automation secret) so it adds hosts, writes check scripts and builds dashboards. NetworkChuck's main AI added the rest of his hosts and built a custom dashboard this way, and also wrote checks for Plex and Home Assistant.
Watch out: He gave his agent an administrator role; Sentinel's least-access approach would give a narrower one.
NetworkChuck's favourite check: each Hermes agent's VM runs a small local script that Checkmk's agent executes. It reports whether her crons are running, whether the gateway is up, how many hours her model login has left, whether she is replying, and usage. He then had his AI build a "Hermes agents" dashboard that showed two agents critical (no backup found, model login issues).
For you: A gateway-up and model-login check per agent would feed the Gauntlet's Fleet and Alerts pages.
A self-hosted uptime monitor. It checks HTTP(s), keywords, JSON answers, TCP, ping, DNS, Docker containers and push heartbeats, and can alert to Discord, Telegram, Slack, email and 90+ others. It can publish status pages. MIT, about 93k stars.
For you: A push monitor could act as a heartbeat each Hermes cron job pings; that would feed Sentinel's channel.
Watch out: Keep its status pages private unless you mean them to be public.
Checked 26 Oct 2026
Several presenters use an Obsidian vault as the readable wiki for themselves and their agents. Wanderloots builds an LLM wiki with Git save points and an "Obsidian Safe" allow-list so agents only touch approved vaults. Tina Huang notes Obsidian Sync is paid and end-to-end encrypted and lets several machines and agents share one vault. Eric Michaud splits the vault into a human side and a machine side.
For you: The Cone of Silence on GitHub already plays this role; the human/machine split matches the archive-folder rule.
Three self-hosted wikis are well known. Wiki.js (Node.js, AGPL-3.0, about 29.1k stars). Docmost (a Notion or Confluence alternative with real-time editing, diagrams, spaces and permissions; AGPL-3.0 core plus enterprise parts; about 22.1k stars). BookStack (simple documentation platform, MIT, about 19.1k stars, now run from Codeberg).
For you: Coolify's one-click list has Wiki.js (as "Wikijs") and Docmost, not BookStack. The vault already does this job, so a web wiki is optional.
Watch out: A second wiki splits notes in two places; keep the vault as the single home unless there is a clear reason.
Checked 26 Oct 2026
A homelab start page with many pages, search, keyboard shortcuts, status checks (HTTP and ping) for each app, widgets and themes. Set up in YAML or its own editor. Built-in login with hashed passwords and a read-only guest role, plus Keycloak single sign-on. MIT, about 26.8k stars.
For you: Its built-in up/down dots give a basic status view without a separate tool.
Checked 26 Oct 2026
A light dashboard (single binary under 20 MB) that pulls feeds into one page: RSS, Reddit, Hacker News, YouTube uploads, weather, market prices, Docker container status and server stats. Set up in YAML. AGPL-3.0, about 37.7k stars.
For you: More of a morning news-and-status page than an app launcher; could sit beside Alfred's briefings.
Watch out: The README does not mention any login of its own, so keep it on the private network.
Checked 26 Oct 2026
A simple app launcher and browser start page with a search bar. "Enhanced apps" show live stats from a service's API. Runs as a Docker image from LinuxServer. MIT, about 9.4k stars.
Watch out: Not in Coolify's one-click list, so it would need a custom Docker Compose.
Checked 26 Oct 2026
A dashboard you arrange by drag and drop, with no YAML editing. It has 80+ integrations, built-in users, groups and permissions, and single sign-on through OIDC or LDAP. Apache-2.0, about 5.2k stars.
For you: Good if Chris wants to rearrange tiles on screen instead of editing files.
Checked 26 Oct 2026
A self-hosted start page for your services, set up in YAML files. It has over 100 service integrations, widgets (weather, search, server stats) and Docker integration that shows container status and can find services through labels. GPL-3.0, about 33.4k GitHub stars.
For you: The closest match to a single intranet page linking n8n, Hermes web UIs, Open WebUI, Qdrant and the Gauntlet.
Watch out: The README says widgets can see personal data, so if it is reachable from an untrusted network it must sit behind a reverse proxy or VPN with login; its Docker examples require HOMEPAGE_ALLOWED_HOSTS.
Checked 26 Oct 2026
Chris already self-hosts most of the parts on Coolify: n8n, Hermes web UIs, Open WebUI, Qdrant and the Gauntlet. The videos suggest putting them behind one private network, adding a monitoring page, and keeping the vault as the wiki. This is a suggestion built from the videos, not a plan he has agreed to.
For you: Sentinel reviews Tailscale and Checkmk before anything is installed.
Watch out: No video shows a single intranet start page that links all of these; that page would have to be built (the Gauntlet could be it).
Install and configure the tools in the videos, in order, with a check at the end of each. Where it touches Chris's own stack, a fit line says what changes.
From your videos: 22 procedures from 37 of the 105 videos. Hermes, VPS hardening, Ollama and MCP have full steps; no video walks through Coolify, Tailscale or n8n self-hosting. Plus 10 entries from the web (marked From the web) and web-sourced steps for the Tailscale and n8n starter-kit entries, filling the Coolify, Tailscale, n8n, Hermes install and update, Obsidian Git, OpenRouter key and SSH gaps.
Coolify's own first-app guide uses a ready-made Docker image (nginx), so there is no code or build to set up. It is a safe way to check a new server works end to end.
For you: Matches Chris's rule for changes: Stop, then Deploy, not Restart. Delete nothing afterwards; archive per his rule.
Watch out: A test app on a public domain is public. Remove or stop it when done, or keep it on Tailscale only.
Checked 26 Oct 2026
Coolify is a self-hosted control panel for apps and services. It needs Linux (Ubuntu LTS is fine; for non-LTS Ubuntu use the manual install), 2 CPU cores, 2 GB RAM and 10 GB free disk, on amd64 or arm64. No video walks through it; these steps are from Coolify's docs.
For you: Chris already runs Coolify on agent86.cloud; this is for a rebuild or a second server. Keep the .env backup out of the vault and out of Discord.
Watch out: Coolify warns that whoever reaches the sign-up page first becomes admin with root access to the server, so do step 5 at once.
Checked 26 Oct 2026
MCP lets an agent use outside tools. Tina Huang adds NotebookLM by asking Hermes; Metics Media shows an MCP catalogue with toggles.
For you: Keep the list short per agent; each connected server adds text the model reads every turn.
Watch out: Sharbel A. says, for Claude Code, that one connected service can add tens of thousands of tokens (his GitHub example: 26,000).
Telegram has a QR flow in the Hermes dashboard. Other channels (Discord, Slack, WhatsApp, email) are listed, and Tina Huang connects them by asking Hermes to help.
For you: Chris already uses Discord, one channel per agent. Set each agent's home channel to its own Discord channel.
Hermes needs a model before it can think. The videos show three ways in: the dashboard Keys page, the hermes model command, or a subscription login.
For you: Chris uses OpenRouter with Claude Haiku: one key per agent, each with its own limit. Keys never go into Discord.
Watch out: If ChatGPT login fails, Python Simplified says to enable device code authorisation in ChatGPT's security settings.
Several videos install skills by pasting a GitHub link into the agent. Others say to review first. This order puts the review before the install.
For you: Chris's rule: nothing is installed without Sentinel's review and his yes.
Watch out: The Next New Thing reported a skill that pushed users to a paid product from inside its own file.
Hermes's own docs give one command for Linux, macOS and WSL2, one for Windows, and desktop app bundles for Mac (Apple Silicon only; Intel Macs are not supported) and Windows. The script needs Git, curl, tar and SHA-256 tools.
For you: Official repo: github.com/NousResearch/hermes-agent. Chris's agents run on Coolify, so this matters for a Mac test install; Sentinel reviews before anything is installed.
Watch out: Hermes needs a model with at least 64K tokens of context (Quickstart page).
Checked 26 Oct 2026
Three videos use Hostinger's Hermes template. Python Simplified adds HTTPS and shows where the login is kept.
For you: Chris runs Hermes under Coolify, not Hostinger's Docker Manager. For changes there: Stop (untick Docker Cleanup), then Deploy, not Restart.
Watch out: Sharbel A. and Metics Media warn about prompt injection; give the agent a machine with nothing else on it.
The desktop app or the terminal installer puts Hermes on a Mac or PC. Tina Huang lists four homes for it: a dedicated machine, a VPS, an old laptop, or your own computer inside Docker.
For you: On Chris's Mac this is best used as a window onto the server agent (see the remote gateway entry), not a second agent.
Watch out: Tina Huang: if it runs on your personal computer, use Docker so it cannot reach the rest of the machine.
Jack Roberts says never paste keys into a conversation, because they stay in history. Lead Gen Jay keeps tokens in one store; Metics Media saves them in a password manager.
For you: Matches Chris's no-secrets-in-Discord rule.
OpenRouter says 'We recommend setting a credit limit on every key.' A key that hits its limit is refused until you raise it; daily, weekly or monthly limits reset on their own. Guardrail caps are set in USD.
For you: Fits Chris's one-key-per-agent cost tracking and the 01 Nov 2026 budget reset. Convert USD limits to CAD at the day's rate when budgeting.
Watch out: OpenRouter's blog: there is no warning before a budget runs out; requests just start failing with a 403. A key with no limit can drain the whole balance.
Checked 26 Oct 2026
The desktop app can open the agent that lives on your server, so you get one agent with two doors instead of two agents. Teku AI also lists SSH and Hermes Cloud gateways.
For you: For Chris this would be the Hermes dashboard address on agent86.cloud.
hermes update pulls the latest stable release, backs up key files first, checks the key startup files compile (and rolls back if not), and restarts running gateways. Image- or package-managed installs (such as Docker) are different: Hermes reports the outside update command instead.
For you: If Chris's Hermes runs from the Docker image under Coolify, the Docker route applies; in Coolify that means Stop, then Deploy (his standing rule).
Watch out: Rolling back is manual: an older release cannot read a state.db the new one has written, so keep the snapshot. Chats made after the update are not in it.
Checked 26 Oct 2026
For testing or teaching, Tina Huang uses a Colab notebook with Google's free T4 GPU.
Watch out: Sessions expire and wipe unsaved work; nothing there is private.
Several presenters use Tailscale, a private network between your devices, so agents can reach other machines or local models without opening them to the internet. None of the videos shows the install clicks.
For you: Fits Chris's rule that dashboards are reached over Tailscale or a tunnel only. Sentinel should review before install.
Watch out: The videos explain why, not how; the steps here come from Tailscale's own docs (From the web). Keys expire after 180 days by default, so see the How-to on key expiry for the server.
Tina Huang ranks Ollama as the easiest local route. Code and agents reach it on port 11434. She also shows Ollama's launch tab offering a Hermes command.
For you: Hermes on the VPS cannot reach a model on Chris's Mac by localhost; Tina Huang names Tailscale for linking local models to a VPS.
Watch out: Sharbel A.: a weak local model on a messy multi-step job can waste more time than it saves.
Tristen O'Brien adds an MCP server to Claude as a custom connector before using Claude Motion with Higgsfield.
Watch out: He says Claude Motion is a beta on Team and Enterprise plans only.
KodeKloud shows the config file most coding tools use: a JSON file with an mcpServers list, each with a command, arguments and optional environment variables.
For you: Sentinel should check authentication and what data leaves for any remote server.
Watch out: He warns that community servers may be untested or out of date.
Anthropic's quickstart installs Claude Code in the terminal and signs in on first run. You need a Claude Pro, Max, Team or Enterprise plan, or a Claude Console account.
For you: Chris already uses Claude Code; useful when setting up a new Mac.
Watch out: The native install updates itself; the Homebrew install does not (run brew upgrade claude-code).
Checked 26 Oct 2026
Pav Rusovs splits it into four layers: Memory, Agent, Pulse and Screen. He chose Claude Code over Hermes for this.
For you: The 'short map plus signposts' idea suits Chris's agents' always-loaded files.
Watch out: Steps are the video's layers from the notes; the folder-sync tool is not named.
RoboNuggets covers mods: add-ons that change Claude Code's interface and behaviour.
NetworkChuck installs the free Community edition on an Ubuntu VM in a few minutes.
For you: Place it where it can see the VPS and home machines (he says the server must be able to reach what it watches). Sentinel reviews before install.
Watch out: Changes only take effect after 'activate pending changes'.
AI Foundations sets up two credentials for a research workflow: an Airtable token scoped to one base, and an OpenRouter key sent as a Bearer header.
For you: Use a separate OpenRouter key for n8n with its own limit.
n8n's docs list a one-line script, Docker Compose, cloud server guides and npm, and say npm is deprecated from n8n 3.0. n8n's docs do not mention Coolify, but Coolify's own docs offer n8n as a one-click service in three versions: n8n alone, with PostgreSQL, and with PostgreSQL plus a worker.
For you: Chris runs n8n on Coolify already. Coolify's three versions: n8n alone with SQLite (testing, low volume); n8n with PostgreSQL 16 (production); n8n with PostgreSQL, a worker and Redis (high volume, parallel runs).
Watch out: Coolify's n8n page gives no step-by-step clicks, so none are listed here. Moving from SQLite to PostgreSQL later is a migration, not a switch.
Checked 26 Oct 2026
Tina Huang names n8n's Self-hosted AI Starter Kit, which bundles n8n, Ollama, Qdrant and Postgres, run with Docker. The notes do not record the install steps.
For you: Chris already runs n8n on Coolify; the kit matters only if Qdrant and a local model join it later.
Watch out: n8n's own README says the kit is not fully optimised for production; it is for trying ideas. On a Mac, Docker cannot use the GPU: run Ollama directly on the Mac and set OLLAMA_HOST to host.docker.internal:11434 in .env.
Tina Huang's fourth memory tier: a vault Hermes can read and write with its Obsidian skill.
For you: Chris's vault 'The Cone of Silence' is already synced to GitHub; Brainiac files into it, and the archive folder stays off limits.
Pieced together from several videos: Git for undo, a rules file at the root, a split between human and agent folders, an index and a log, and the command-line setting agents need.
For you: Put Chris's rule in the root rules file: no agent reads the archive folder, and agents archive instead of deleting.
Watch out: Mason Fitzgerald and Simon Scrapes say an instruction file is soft guidance; put must-never rules in permissions or hooks too.
Obsidian Git is a community plugin that can commit, pull and push your vault on a schedule and pull when Obsidian opens. Its own docs call the mobile version very unstable.
For you: Chris's vault already flows iCloud to GitHub; this documents the plugin route for reference, not a change. Keep the token out of the vault and Discord.
Watch out: The author advises against using it on mobile (crashes on clone or pull, no SSH); on iOS do not store the vault in iCloud if you do try it.
Checked 26 Oct 2026
Metics Media's beginner procedure on a plain Ubuntu LTS server: update, make a normal admin user, log in with an SSH key, and turn on a firewall. Every command below is read out in the video.
For you: Chris's VPS already runs Coolify: before enabling a firewall, list the ports Coolify and its apps need, or you can lock services out. Sentinel should review firewall changes.
Watch out: The video does not show turning off root or password login; do not assume it is done.
Coolify's firewall page lists six ports for a server running Coolify itself. Once the dashboard works through a domain, the three direct-access ports can be closed, because everything then goes through 80 and 443.
For you: Answers the 'list the ports first' note on 'Harden a new Ubuntu VPS'. Chris's dashboard is already on coolify.agent86.cloud, so 8000, 6001 and 6002 should not need to be public.
Watch out: Coolify: 'Do not open ports that your server does not use.'
Checked 26 Oct 2026
The video procedure stops before this. Ubuntu's docs say a settings file in /etc/ssh/sshd_config.d/ wins over the main file, because SSH uses the first value it reads. Tecmint names the three lines to set and warns that cloud images can ship a file that turns passwords back on.
For you: Fixes the gap in 'Harden a new Ubuntu VPS'. Coolify connects to its own server over SSH, so check Coolify still shows the server as reachable afterwards; Sentinel should review first.
Watch out: Ubuntu's docs: a mistake here can lock you out of a server you only reach by SSH. Keep Hostinger's browser terminal as a way back in.
Checked 26 Oct 2026
Metics Media points to the provider's backup page as a day-one step. Hostinger defaults to weekly backups, with daily as a paid option.
For you: Take a snapshot before Coolify or Hermes upgrades.
GitHub repos and skill packs the videos recommend, grouped by what they do. Free replacements for paid apps are on the next tab.
From your videos: Fed by 31 of the 105 videos, mostly the GitHub roundup and skills videos. Many videos give no owner/repo; those are marked. Plus 8 entries from the web to fill gaps (marked From the web), and real GitHub links added for 23 entries the videos left without an owner. Updated 26 Oct 2026 after reading three more full transcripts (IbFaY3xFpZM, BrBQRNIDRS8, x2F3fdn82GU): 17 more entries.
A library of 279 role-based agents across company divisions. AI Edge cites about 152,000 stars and 124 contributors.
For you: A place to borrow role ideas for the seven agents, not something to install whole.
Watch out: A giant pack: load only the one role you need, not all 279.
Asks the same question 100 times, then pits the answers against each other in a bracket until one wins. At the end it shows why the winner beat each rival. Jack Roberts says it is new and mostly talked about on Instagram.
For you: Cerebra, for one hard question where the best answer matters.
Watch out: 100 answers means roughly 100 times the tokens of one answer. The repo is from Jack Roberts's video description.
Connects an agent to your real Chrome browser and repairs itself when a step fails. Listed in Teku AI's Hermes skills video.
For you: Cerebra.
Watch out: It drives your real, signed-in browser, so Sentinel should review it first.
A free browser built for AI agents. The agent browses in its own tabs, so it does not take over your screen, and you can bring any model. Jack Roberts shows it scraping posts from an X account.
For you: Cerebra, for web research without seizing Chris's own browser.
Watch out: The real name is Ego Lite (site lite.ego.app). macOS and Windows only for now; Linux is on its roadmap.
The agent framework Chris's seven agents run on, from Nous Research. The README calls it self-improving: it creates skills from experience and improves them during use. It works with OpenRouter and many other model providers, and has a terminal interface plus Hermes Desktop.
For you: The source of truth for every Hermes question; check its docs before trusting a video's claim.
Watch out: Read release notes before upgrading the Coolify deploy; plugins like Phone Harness need specific versions.
The official collection of reference MCP servers (MCP is the plug standard agents use to reach tools). The README points to the MCP Registry for the full list of published servers.
For you: A trusted starting point before any third-party MCP server; Sentinel can compare against these.
Watch out: The README says these are reference examples to learn from, not hardened production servers.
Sharbel A.'s own dashboard for running many Hermes agents: send out tasks, health checks and live cost tracking.
For you: Ideas for Chris's planned agent dashboard (fleet status and cost per OpenRouter key).
Watch out: Presenter's own tool; no owner/repo stated.
Bundles n8n with Ollama (local models), Qdrant (a vector database) and Postgres in one install. Tina Huang names it in her open-source AI overview.
For you: Close to Chris's own n8n plus planned Qdrant setup; useful as a reference layout.
Splits one job across specialist agents or command-line workers, runs them side by side, then checks the result. Sharbel A. lists it among his Hermes skills.
For you: Forge, for big coding jobs that can be split up.
Watch out: The captions garble the name; the catalogue gives the repo. More agents in parallel means more tokens.
Dubibubi's number 3 skill. It splits a big task into parts, hands them to specialist agents or outside CLI workers (Codex, Gemini, Cursor), runs them side by side, then verifies the result instead of stopping half done. It is modelled on oh-my-claudecode. The video claims up to 50% token savings through smart routing. EveryDev lists an Oh My Hermes plugin by Hope Kim (@rlaope), MIT licence, with 116 skills.
For you: A cure for the yes-man problem on big jobs; test on one agent first.
Watch out: The video gave no owner, and more than one repo uses this name. Confirm the owner before Sentinel reviews it.
An open-source coding agent for the terminal, like Claude Code but free to change and able to use any model. It also has a VS Code plugin. The Next New Thing cites 220,000 stars; one host's venture studio builds its own agent harnesses on top of it.
For you: Cyborg (code) could compare it with Claude Code; Sentinel reviews first.
Watch out: OpenCode's cloud service was among services hacked that week, so cloud users should rotate API keys. The hosts say the local CLI repo itself was not compromised.
Open-source agent video production with 12 pipelines and more than 700 skill files. Named in two Hermes skill lists and inside Julian Goldie's Agent OS.
For you: Forge, if video becomes a real job.
Watch out: 700+ skill files: a skill pile that big costs tokens if loaded whole.
Runs several coding agents side by side in one window. You hand each agent its own task, they work at the same time, then you review each result side by side. Jack Roberts cites about 89,000 stars, most of them this month.
For you: Forge or Cyborg, if Chris ever runs parallel coding agents on the Mac.
Watch out: Several agents at once means several times the tokens. The README also offers a phone companion app.
A coding harness that turns several models into a dev team: one plans, one builds, one reviews. A manager role picks which model does which task to save money. Each role gets its own instructions and MCP connections. The Next New Thing cites 78,000 stars.
For you: Same idea as Chris's agent lanes, but for code; Cyborg's area.
Watch out: The captions spell it 'Roof low'; a web search shows Ruflo is a multi-agent tool for Claude Code and Codex, so this is likely it. Check the repo before installing.
A free command-line tool that reads Claude Code's local log files and reports token use and cost by day, month and session. It can also feed a status line.
For you: Pairs with Codeburn for Chris's planned cost-per-key dashboard.
Shows where your Claude tokens actually go and flags waste, such as an MCP server you never use but that loads on every message. Nate Herk points to it right after showing the token-saver repos fell short.
For you: Forge (cost watching).
Watch out: Needs Node.js 22.13 or newer (per its README). It reads local session files only; no account.
A toolbox that connects agents to 1,000+ apps such as Gmail, Slack, Notion and HubSpot through sign-in, also as a CLI.
For you: Alfred.
Watch out: Each connection is account access; scope it per agent.
Dagger's sandboxes that run each agent in its own container.
For you: Sentinel would like this: agent code stays boxed in.
The NSA's free decompiler, now driven by agents. The Next New Thing had an agent decompile a test app and pull out its hidden password. It is one of the three decompilers REA drives.
For you: Cyborg's cyber work, on test apps only.
Watch out: Hard to use by hand; the guest says nobody learns it from scratch now, they let agents drive it.
Lets any agent work in Drive, Gmail, Calendar, Docs, Sheets and Slides, with 40+ skills. Nate Herk says it covers more than the Claude desktop Google connector and works in Claude Code, Codex or your own agent.
For you: Alfred (reminders and briefings).
Opens a real browser and clicks through your app like a customer, then writes tests you can rerun. Because it is a command-line tool, not an MCP, it does not dump the whole page into Claude's context, so it is lighter on tokens.
For you: Cyborg and Forge, before anything ships.
Microsoft's official browser automation through accessibility snapshots.
Watch out: Nate Herk prefers the CLI version because the MCP loads more into context.
Has an agent study how an app works, from its behaviour down to its binaries, and gives that evidence to your coding agent so it can rebuild a feature. Claimed 240,000 to 300,000 stars. Uses include drivers for old hardware and a Stream Deck on Linux.
For you: Cyborg, for one's own devices (for example sim-racing hardware).
Watch out: Copying someone's software to sell breaks terms of service (Nuno Tavares; Adobe's terms ban it). The Next New Thing says basic models already did about as well in one test.
Skills, tools and a shared knowledge base that let a coding agent mod PC games you own: it finds the engine, reads the code, builds the mod and tests it in the running game. Listed in Jack Roberts's video description.
For you: Cyborg, for sim-racing tinkering on a PC, if ever wanted.
Watch out: Game mods can break a game's terms or anti-cheat in online play; keep it to offline games.
Builds a command-line bridge between an agent and a desktop app that has no MCP, so the agent can open it and change settings without clicking. It has a marketplace of ready-made ones, Blender among them. One host's agent used it on its own to drive apps it had installed.
For you: Forge, for desktop editing apps; Sentinel reviews each bridge.
Checks your hardware and scores which local models will fit, by memory, speed, quality and context size, before you download one.
For you: Useful before trying any local model on the Mac.
VisiData (spreadsheet-style view of CSV files in the terminal), croc (end-to-end encrypted, resumable file transfer between computers), kondo (finds build junk such as node_modules in old projects), dnsglobe (watches DNS changes spread across 34 public resolvers on a world map), gitlogue (replays Git history as an animation in the terminal) and grex (writes a regex from example strings).
Watch out: Names confirmed from the video's description: dnsglobe, gitlogue and grex.
A terminal screen to see and manage Docker containers: status, resource use and logs in one place, no copying container IDs.
For you: Forge, for the Coolify Docker side of the VPS.
Makes Git worktrees (separate folders for separate branches) easy to create, switch and remove. The video notes it suits running Claude Code agents in isolated worktrees.
For you: Cyborg, for parallel coding jobs.
Reads, edits, merges and converts YAML, JSON, XML, CSV and TOML from the command line, handy inside scripts.
For you: Forge, for config files.
A large library of design components (headers, footers, pricing blocks, hover effects). Its MCP lets Claude pick components itself instead of you browsing. Jack Roberts rebuilt his Glido app's UI with it.
Watch out: The old Magic MCP has been replaced by the '21st MCP', set up through the 21st CLI (per its README).
Plain English to editable HTML architecture and workflow diagrams. AI Edge cites 64,400 stars.
For you: Could draw the agent86.cloud stack as a living diagram.
Watch out: Spelling confirmed. Hermes also lists Archify as an optional skill; Sentinel still reviews it.
Turns one sentence into a designed diagram, in about 45 types (flows, quadrants, timelines, org charts, charts). The Next New Thing says it is by designer Katherine Lavery and has about 55,000 stars; Jack Roberts uses it for flowcharts.
For you: Useful for vault notes and Chris's visual learning style.
Watch out: Jack Roberts: the same prompt can give slightly different results. The author is Cathryn Lavery (captions said Katherine).
Nate Herk's own repo (he discloses it): a scroll engine plus design rules for Apple-style scrolling websites, with a checker that screenshots the scroll so Claude fixes problems first.
Watch out: Presenter's own product; link is in his description, not stated in the video.
200 free moving backgrounds (gradients, glass, galaxy effects) for any website. Nuno Tavares calls it a hidden gem that works well with Claude Design.
Watch out: Few stars yet; no owner/repo stated.
Checks English writing for common AI writing patterns, scores the text in five categories and gives editing instructions. Use it on a README, landing page, email or script. Listed in Jack Roberts's video description.
For you: Forge, as a final pass on anything written for other people.
Watch out: It finds patterns, not facts; a clean score does not mean the text is right.
Describe a part and get an STL file for 3D printing. The Next New Thing printed a phone stand from one prompt; the skill turned off telemetry because the readme suggests it.
Watch out: Nuno Tavares: blocked by default on a fresh Windows 11. Mac not tested in these videos.
Indexes a codebase into a lasting knowledge graph, so the agent explores less. Sharbel A. repeats a claim of 99% fewer tokens; Teku AI says 120 times fewer; about 45,000 stars.
For you: Forge and Cyborg.
Watch out: The savings figure is the project's own claim; measure it before trusting it.
Persistent agent memory with three steps: retain, recall and reflect (it looks for patterns, not just stores). Eric Michaud deployed it in Docker on his Hostinger VPS and set it as Hermes's memory with the 'hermes memory setup' command. Nuno Tavares calls it 'level three memory'.
For you: Brainiac or Cerebra; same VPS-plus-Hermes setup as Chris.
Watch out: Hermes allows one external memory provider at a time, and it overlaps Chris's Qdrant plan.
An external memory provider shared across agents, with 'dreaming'. Nuno Tavares says his team uses it a lot. Check it with 'hermes honcho status'.
For you: Brainiac.
Watch out: Pick Honcho or Hindsight or Qdrant; not two at once.
Shared memory across Hermes, Claude Code and Codex through MCP, with a free tier. Sharbel A. uses it in his workflows (captions say 'Memola').
For you: Brainiac.
A skill pack by Obsidian's CEO that teaches agents the Obsidian CLI, markdown, Bases, Canvas and Defuddle.
For you: Brainiac (vault filing) in The Cone of Silence.
Keeps task_plan.md, findings.md and progress.md, with a hook that keeps the plan in the agent's context. AI Labs lists it first for Hermes.
For you: Forge, for long multi-step jobs.
Qdrant's own MCP server. It gives an agent two tools: qdrant-store (save a piece of information, with optional metadata) and qdrant-find (search for it later by meaning).
For you: Directly fits Chris's Qdrant vault-memory plan, starting with Cerebra.
Watch out: Pick one memory system per agent: Qdrant, Honcho or Hindsight, not two at once.
A trending repo (about 12,000 stars) that keeps one set of memory files for every coding tool you use: Claude Code, Codex, Cursor, Grok, Devin. Any agent on any machine that can read the files knows what the others did. One host lost a conversation because it was in Codex on his other computer.
For you: Same idea as Chris's Obsidian vault acting as shared memory.
Watch out: The captions call it just 'AI memory'; no owner given.
The Next New Thing also names Cortex Suite (agent memory and codebase analyser, works with Graphify), Smriti Brain (local-first project memory for coding agents) and a YouTube subscriptions ingest repo that turns videos into markdown using Karpathy's LLM-wiki idea.
For you: The YouTube ingest idea is close to how this toolbox was built.
Watch out: Named briefly only; no owner/repo stated.
Two Claude Code plugins Nate Herk recommended in May and now says to hold off on.
Watch out: Claude Mem sends usage data by default; Context Mode has an open bug that can burn through your usage.
Caveman (JuliusBrussee/caveman) claims about 65% fewer output tokens; RTK (rtk-ai/rtk) filters terminal output; Ponytail is another saver. Nate Herk cites JetBrains tests: Caveman cut about 8.5%, RTK saved nothing and at low effort cost more, Ponytail saved about 10%.
Watch out: Ponytail has an open Windows bug that can stall each message. Measure before trusting any saving.
Gives agents read access to X, Reddit, YouTube, GitHub, Instagram and more, with no API fees and a fallback when one source fails. Teku AI cites about 81,000 stars.
For you: Cerebra (web research).
Watch out: Teku AI praises it for getting past 403 bot blocks; the warnings page says let Sentinel decide whether a bypass tool is acceptable.
Connects Claude to thousands of ready-made scrapers for Instagram, TikTok, Amazon and more, where Claude would otherwise get an empty page. Nate Herk shows it pulling Chicago bakeries from Instagram into a spreadsheet.
For you: Cerebra.
Watch out: The connector is open source but the scrapers run on Apify's cloud: $5 of free credit a month, then pay per result.
Karpathy's loop that runs experiments overnight.
Watch out: It trains a small language model in 5-minute runs, so it needs a real GPU; it is a research toy, not an office tool.
Feeds the agent current library documentation, so it codes against today's version.
For you: Forge and Cyborg.
Strips a web page down to clean reader-mode markdown, so the agent reads fewer tokens. By Obsidian's CEO.
For you: Cerebra and Brainiac: cheaper reading before filing into the vault.
An autonomous agent that writes research reports. Listed in Dubibubi's skills roundup.
For you: Cerebra.
Researches a topic across Reddit, X, YouTube, Hacker News, Polymarket, GitHub and the web over the last 30 days. Futurepedia lists it first; AI Edge notes 140 contributors.
For you: Cerebra.
Andrej Karpathy's repo where several models, each with a persona, debate to improve a decision.
For you: Cerebra, for a second opinion on a big decision.
Credited in the video to Matt Berman. A loop tells the agent what to do with a result: measure it, keep the change only if it helps, repeat until a target, stop when it stops improving. The skill gives a catalogue of ready loops and helps design new ones in plain language.
For you: Fits Forge or Cyborg for 'make this faster' style jobs.
Watch out: Honourable mention only; no repo given.
Adds deepfake checks to anything your agent takes in: audio, images, video and text. It tries to name the tool that made it (for example ElevenLabs, ChatGPT or Claude) and labels it as AI for the agent.
For you: Cerebra, when research pulls in a lot of web content.
Watch out: No repo or price given in the video.
Replaces Hermes's built-in YouTube skill, which breaks on a VPS because YouTube blocks cloud IP addresses. It does transcripts, channel browsing, playlists and video search with no Google API key. Dubibubi says it runs on an API that handles 15 million transcripts a month.
For you: Directly relevant: Chris's Hermes runs on a Hostinger VPS. Cerebra (research).
Watch out: It depends on an outside transcript service the video did not name. Also, this very video's transcript contains a sentence aimed at AI agents reading it, so treat every transcript as untrusted text.
Anthropic's security plugin sends a team of agents through your code looking for holes; separate agents try to prove each other wrong. You get a plain-English report of what survived, and it can write the patches.
For you: Sentinel (security review), on code only.
Watch out: Anthropic lists the Claude Code plugin as beta; the full Claude Security app is a public beta for Enterprise plans only.
Turns a coding agent into a security auditor: isolated agents run reconnaissance, coverage, hunting, candidate validation and independent verification. Jack Roberts says run it before shipping any client app.
For you: Sentinel.
Watch out: Many isolated agents means many tokens per audit; run it on one repo at a time.
Red-team and regression tests for prompts.
For you: Sentinel, to test agents' prompts against injection.
A viewer project offers a Linux sandbox where the AI cannot break your laptop. The host's point: telling an AI to stay in a folder is not enough. A real container is a wall it cannot get past, which protects files and customer data.
For you: Chris's agents already run in Docker under Coolify, which is this idea.
Watch out: The viewer project had no stars yet; Container Use (Dagger) is the better-known option.
Security-scans an agent skill before you install it. Futurepedia shows it.
For you: Sentinel's first check on any new skill.
Watch out: Skill Inspector is a skill inside NVIDIA's SkillSpector scanner. Its optional AI pass needs an OpenAI-compatible key.
About 24 to 25 senior-engineer skills: /spec writes a real spec with you, /plan breaks it into small tasks and /build works through them one at a time with a check on each. Nate Herk says it keeps the plan in files, so a messy session can restart cleanly, and that it is easier for beginners than Superpowers.
For you: Forge or Cyborg, for builds that run longer than one session.
Official skills that teach coding agents to build Android apps well: battery use, background behaviour and data loading. The hosts' favourite is the on-device AI skill, which uses the ML Kit GenAI Prompt API to talk to Gemini Nano on the phone. Android's own page says the skills live in a GitHub repo and can be installed with the Android CLI or used in Android Studio.
For you: Not for Chris's Apple devices; the iOS equivalent is the SwiftUI agent skill.
Dubibubi's number 4: over 700 security skills mapped to MITRE ATT&CK (a public list of attacker methods). It covers threat modelling, vulnerability checks, secure code review and incident playbooks. Despite the name, it was built by researcher mukul975, not by Anthropic. One web listing counts 754 skills under Apache 2.0, also mapped to NIST frameworks.
For you: Sentinel could borrow one or two skills for reviews.
Watch out: A giant pack: the warnings list says not to load packs like this wholesale. Pick only the skills Sentinel needs.
A hand-picked, regularly updated list of Claude Code resources: skills, hooks, slash commands, status lines and tools. Useful for finding the real repo behind a name heard in a video.
For you: A lookup list for Chris and Sentinel, not something to install.
Watch out: A list is not a review; each linked item still needs Sentinel's check.
A small viewer-made skill set that makes a coding agent ask when it does not know, instead of guessing. It had its first GitHub star on the show.
Watch out: Brand-new, one-person project; no repo owner given.
Cursor's plugin pack connects 90-plus apps and bundles skills. The hosts point to a full plugin pack by a creator heard as 'P Stack' by 'potato'. Its /ketchup skill catches you up on what changed in a chat since your last message, in plain words, and lists every decision waiting on you.
For you: The /ketchup idea suits Chris's wish for short catch-ups; it could be copied as a Hermes skill.
Watch out: Names are garbled in the captions; no owner or repo given.
A pack of 292 skills that runs plan, test, implement, review and verify, with remember and improve steps. Named in The Next New Thing's top-repos roundup.
Watch out: Giant pack: the README now lists 71 agents and 311 skills. The warnings page says not to load packs wholesale.
Expo's own skills for building mobile apps, from idea to an app on your own iPhone in an afternoon. They include Expo's list of 20 things that make an app look AI-built and the current App Store rules.
For you: Cyborg (iOS work).
Watch out: You need a $99-a-year Apple developer account for TestFlight or the App Store.
Garry Tan's chain of role skills for shipping software: /office-hours questions the idea, /plan-ceo-review reviews the plan, /codex gets a second opinion, plus 'careful' guard rails and iPhone QA. AI Edge cites 133,000 stars.
For you: Cyborg, for iOS work (it has an iPhone QA step).
Dubibubi's number 2 skill turns a popular article on small UI details into a skill. Once installed, the agent applies them on every UI: headlines that don't leave one word alone on a line, matching corner radius for a button inside a card, icon animations, tabular numbers so stats don't jump, and animations that can be interrupted. The GitHub page shows an MIT licence and about 3,800 stars.
For you: Useful for the AI Toolbox page and the planned agent dashboard.
About 27 small skills that make Claude work like a senior engineer. Grill Me asks about 20 questions before acting, Handoff writes a short note so a fresh chat can pick up where the last one stopped, and TDD writes tests first. The Next New Thing calls them battle-tested; Jack Roberts says it is the most-starred skill pack on GitHub.
For you: Grill Me suits every agent; Handoff fits moving work between Claude, Codex and Hermes.
NVIDIA's official skills for Claude Code, Codex and other coding agents. Skills here pass NVIDIA's SkillSpector scan and review before they are published and signed.
For you: A safer skill source for Sentinel to approve from than random registries.
Watch out: Mostly NVIDIA-stack topics (GPUs, CUDA and similar); install only what an agent needs.
Anthropic's reference skills: PDF, Excel, PowerPoint and Word handling, frontend-design, skill-creator, brand guidelines, canvas design, doc co-authoring and web artifacts builder. Dubibubi walks through several of them.
For you: A trusted author, which matters for Sentinel's review.
Skills for people in the sciences, such as adding statistics or physics to a chemistry report, or making research posters in LaTeX (a typesetting language).
Watch out: One skill file told agents to run every session and push the maintainer's paid product, K-Dense Web. Three GitHub issues complained. Proof that a skill file can carry an ad or a hidden push.
Dubibubi says to install it first. After each session it runs an 'evolution loop': it reviews the skills the agent used, merges overlaps, rewrites weak ones and updates the skill library on its own. Over 2,000 stars in the video. Hermes Atlas says it works with Hermes, Codex, Claude Code and OpenClaw, and can share one skill library across agents and devices.
For you: Interesting for a seven-agent fleet that shares skills.
Watch out: It rewrites skills without asking, which skips Chris's rule that Sentinel reviews every change. Keep any validation step on, or skip it.
Official skills that teach Claude each company's security rules, such as users only seeing their own data and checking a payment really came from Stripe.
Watch out: Nate Herk says the full Stripe plugin may route every vendor decision through Stripe.
A methodology of 20+ skills for brainstorm, plan, test-first, implement, review and debug. Nate Herk still rates it, Teku AI cites about 287,000 stars, and it also ships a Hermes plugin.
For you: Forge. Pick it or Addy Osmani's pack, not both.
Watch out: Nate Herk says Addy Osmani's pack is easier if you are new.
Gives coding agents current Swift and SwiftUI documentation. The Next New Thing's guest says Apple's docs are out of date, so agents keep failing on Swift; this feeds them current changes.
For you: Cyborg, for iOS and Swift work.
Watch out: No owner/repo stated.
A skill that finds the best skill for a task. Jack Roberts says it is the top skill on skills.sh.
For you: Sentinel could use it to shortlist, then review before anything installs.
Watch out: Jack Roberts: never install a skill without having Claude or ChatGPT check it first.
From HeyGen. Claude writes an HTML page with animations and HyperFrames renders a real MP4 on your computer: captions, title cards, promos, cartoons. Nate Herk prefers it to Remotion; it is open source with no per-render fees.
For you: Forge (video editing).
Watch out: Needs Node.js 22. Nuno Tavares says some viewers find the style nauseating.
Give it a reference video and it rebuilds the 'recipe': shots, captions, B-roll, effects and timing, so you can reuse a format, including your own best video.
Watch out: Nuno Tavares says skip it: the face, voice and video parts need paid AI services. Eric Michaud warns against copying others one-to-one.
Open-source video made from React code, with a built-in map-animation skill. Tristen O'Brien calls it the free-tier alternative to Claude Motion.
For you: Forge. Pick it or HyperFrames.
Free or open-source projects the videos pitch as swaps for paid apps. Each entry starts with the paid app it replaces.
From your videos: Fed by 18 of the 105 videos. Most claims come from presenters' own tests; few give an owner/repo, and several apps are early alpha. Plus 6 entries from the web to fill gaps for Notion, 1Password, Acrobat, Zapier, cloud storage and paid LLM APIs (marked From the web), and real links added for 25 entries. Updated 26 Oct 2026: 3 entries from full transcripts of x2F3fdn82GU and BrBQRNIDRS8.
Replaces: Firecrawl (second choice). Mentioned briefly as another open-source scraper.
Watch out: Mentioned only in passing in the video.
Replaces: Jev, a cheap paid decision model. A small model that runs free on your own computer and sorts text into choices, such as spam or not, in about 26 milliseconds.
Watch out: Untrained it scored 36% against Jev's 73% (Jack Roberts). Nuno Tavares got one wrong in his test. The real name is Laya (captions said 'Layer' and 'Leia'); its README gives about 33 ms per question on a T4 GPU.
Replaces: paid LLM APIs. Chains about 635 free LLM providers, with a desktop and an Android app.
Watch out: Nuno Tavares says skip it: its own readme says 'not for real work' and you land on weaker models.
Replaces: paid LLM APIs for small jobs. Runs open models on your own computer, with installers for macOS, Windows and Linux. Check fit first with llmfit (Repos tab).
For you: Could run a small local model for private work such as Mantis, if the Mac has the memory.
Watch out: Local models are weaker than Claude for hard tasks; test on real jobs before switching anything.
Replaces: Firecrawl. A Python scraper with a stealth fetcher, offered as a free Firecrawl swap and as an optional Hermes skill. Claimed 87,000 stars.
Watch out: Nuno Tavares says skip it: it is built to get past website blockers. The warnings page agrees.
Replaces: Adobe Photoshop, Lightroom, Premiere Pro and more, plus Pro Tools and AutoCAD. A free, open-source set of 12 apps in one launcher, written in Rust: PhotoCraft, LightCraft, FilmCraft, VectorCraft, PDFCraft, EffectCraft, SoundCraft, CAD and more. It has a built-in MCP, so Claude can run the apps for you; Jack Roberts shows Claude editing an image inside it.
For you: Forge (photo, video and PDF work) could test PDFCraft or PhotoCraft first.
Watch out: Mostly early alpha. Eric Michaud says PhotoCraft is furthest along and SoundCraft is laid out like Pro Tools. Site given: getartcraft.com/apps.
Replaces: Adobe Photoshop. Jack Roberts calls it the second free Photoshop rebuild this month, with code inside the layers. He thinks its interface may be nicer than PhotoCraft's.
Watch out: Built as an Xcode project, so it is a Mac app. The repo is from Jack Roberts's video description.
Replaces: Adobe Creative Cloud (he pays about $70 a month). A 'clean room' build: no Adobe code; it was written by Claude in Rust from public file specs and how the apps behave. PhotoCraft is 80 MB against Photoshop's 6.4 GB, and FilmCraft is 90 MB against Premiere's 6.8 GB. PhotoCraft opens layered PSD files. VectorCraft has 89 of Illustrator's 92 tools and works with SVG and PDF. FilmCraft cannot open Premiere projects but imports timelines by XML. EffectCraft runs After Effects scripts but cannot open .aep files.
For you: Fine for learning and personal work, he says; not yet for paid client deadlines.
Watch out: Artcraft itself lists the gaps: generative AI features, 15 missing tools, typography depth and plugin support. VectorCraft struggles with .ai files and logo tracing. Slow on older Macs. It makes money by selling credits for AI models.
Replaces: Adobe Acrobat. An open-source PDF platform with 50+ tools: edit, merge, split, sign, redact, convert, OCR and compress. Runs as a desktop app, in the browser, or self-hosted with a private API.
For you: Forge (PDF work), self-hosted on Coolify with the private API.
Replaces: iCloud Photos. Self-hosted photo backup, part of Switch and Click's day-one self-hosting stack.
Watch out: Chris's Mac, iPhone and iPad already sync through iCloud; this would be a big change.
Replaces: Plex. Jellyfin is an open-source media server; music clients include Finamp (free), Symfonium (Android, $6 once) and Feishin. Navidrome is a music server that is slower to set up.
Watch out: Switch and Click calls Plex the easier choice.
Replaces: big-tech book, audiobook and document clouds (none named). Kavita serves ebooks and comics, Audiobookshelf serves audiobooks, and Papra stores documents with tagging rules and full-text search.
Watch out: The video's own description names Papra, not Paperless-ngx (captions said 'Papr'). Paperless-ngx is a bigger, older option for the same job.
Replaces: Google Drive, Dropbox and similar cloud storage. A self-hosted file sync and sharing server with apps for phones and computers.
Watch out: Chris's devices already sync through iCloud and GitHub; this would be a big change, not a quick swap.
Replaces: cloud note sync for Org-mode notes. bashbunni self-hosts the Organice web client behind Nginx with a WebDAV backend, so her notes reach her phone.
For you: Chris's notes live in Obsidian, so this is for reference only.
Replaces: cloud storage subscriptions (none named) and Parsec. bashbunni shares folders as a network drive with Taildrive (a Tailscale beta) and streams her desktop with Sunshine and Moonlight, all over Tailscale.
Watch out: She rejected Parsec because it has no LAN-only mode or Tailscale support.
Replaces: 1Password. Vaultwarden is a lightweight, self-hosted server that works with the official Bitwarden apps and browser extensions. Bitwarden itself is also open source and was Switch and Click's sponsor.
For you: A home for agent API keys and passwords, in line with the no-secrets-in-Discord rule.
Watch out: Its README says it is not associated with Bitwarden, Inc. A self-hosted password vault must be backed up and kept off the open internet.
Replaces: paid file-compression tools (none named). Optimises and converts any file you copy to the clipboard, with automated pipelines (resize, convert, watermark). It exposes an MCP, so Claude can set up pipelines for you.
For you: Forge (photo, video and PDF work).
Watch out: Source is GPLv3 but the app is freemium. The repo now says public commits are paused and work continues privately.
Replaces: paid folder-rule apps (none named). Watches a folder and runs if-then rules: rename, tag, colour-label and move files. It can read text in images (OCR). No AI.
For you: A no-AI, no-token way to tidy files on the Mac.
Watch out: Its README calls it the open-source Hazel alternative. Still early (alpha and beta releases).
Replaces: macOS Archive Utility and unarchivers such as The Unarchiver and Keka. Browse inside archives, even nested ones, and pull out single files, including from Quick Look.
Replaces: Knock (tap gestures) and rigid posture apps. MacTap uses the laptop's motion sensor so taps on the sides run actions such as copy and paste. Headjust uses AirPods motion data to report how you sat over a session.
Watch out: Snazzy Labs: MacTap misfires on resonant desks; Headjust is buggy and can drift. Headjust's listing does not say it is open source; MacTap's repo was not found.
Replaces: Rufus (which is Windows-only). Makes bootable USB installers on a Mac, checks checksums, downloads macOS installers and writes raw images to drives.
Replaces: the $20 Mole app, CleanMyMac, DaisyDisk, AppCleaner and Pearcleaner. The free open-source command-line version (type 'mo') frees disk space, fully uninstalls apps, optimises the system and shows what fills your disk. Snazzy Labs says none of the others are as thorough.
Watch out: Snazzy Labs calls it slow and a bit dated.
Replaces: paid project-template apps (none named; the host says others exist but are not simple). Copies a template folder, fills in placeholders such as date and topic, and creates linked files every time you start a project.
Watch out: Free on the Mac App Store, but its listing does not say it is open source. It also has an MCP so agents can deploy templates.
Replaces: paid teleprompter apps (none named). A free open-source teleprompter in the notch near the camera, with voice tracking and PowerPoint notes import. Works from plain text files.
Replaces: caffeinate and Amphetamine, plus many small menu-bar utilities. Keeps the Mac awake, sets per-app volume and audio devices, shows hardware, network, disk and power data, and has clipboard history: 73 features in one free open-source app.
Watch out: Name confirmed: Vorssaint. You install only the features you use; setup asks only for the permissions those need.
Replaces: workout apps that charge a few dollars or show ads. The Next New Thing lists an open-source gym tracker with 5,600 exercises built in and about 12,300 stars.
Watch out: The app's name was not said in the video.
Replaces: expensive dental software. A man (named in the captions as Danny Dovans) was told his 3D jaw scan could only be opened in paid dental software. He built a viewer with Claude, called Voxel Viewer, and put it on GitHub as open source.
For you: The pattern matters more than the app: build a small viewer for a file you own.
Watch out: No repo owner given; the name is from auto-captions.
Replaces: Zapier. Activepieces calls itself an open-source replacement for Zapier, with building blocks called 'pieces'. Chris already runs n8n, which does the same job.
For you: Stay with n8n; Activepieces is only worth a look if an n8n flow proves awkward.
Watch out: Two automation tools means two places for workflows to break; do not run both without a reason.
Replaces: Notion (and Miro). Its README calls it a privacy-focused, local-first, open-source alternative for Notion and Miro: docs, whiteboards and planning in one app.
For you: Only if Chris ever wants a shared workspace beyond Obsidian; the vault already covers notes.
Watch out: Another notes app splits knowledge across two places; Obsidian stays the main vault.
Replaces: Notion. Plain markdown-file memory with wiki links that agents can read and write.
For you: Chris's Obsidian vault already does this job.
Watch out: The hosted plan costs $15 a month; self-hosting is the free route.
Replaces: Slack. A Slack-style workspace from Block (the Square and Cash App company) where agents are members next to people. Free and self-hostable, for example on a VPS. Claimed 35,000 stars.
For you: Chris already uses Discord channels per agent, so this is watch-only.
Watch out: Nuno Tavares: still a work in progress; the early Windows build triggers an install warning.
Replaces: Superhuman. An open-source email app with an MCP so agents can work the inbox.
For you: Alfred, drafts only.
Watch out: Hosted plan is $18. Warnings page: agents should draft, never send, email.
Replaces: Beehiiv. Marc Lou built his own open-source newsletter tool on GitHub because he was tired of paying for Beehiiv. You can host it yourself or pay on his site. It is built so agents can work with it directly. One host prefers Loops, a paid service that does marketing, newsletter and transactional email.
Watch out: The captions spell its name like 'Mailchimp', which is a different company. The hosts said it had only dozens of users and one would not use it yet.
Replaces: Semrush and Ahrefs. Plugs Claude into real keyword, ranking and backlink data, so it names the exact page and keyword to fix. Pay per use.
Watch out: Nate Herk: free trial credits are small, and your data sits on their server unless you host it yourself. The README lists the hosted plan at $10 a month.
Replaces: Jira. An agent-first project manager: agents become employees in an org chart, with tasks, Kanban, routines, secrets and board approvals. NetworkChuck runs Hermes, Claude Code, Codex and Pi agents in it. Claimed 100,000+ stars.
For you: Could sit above Chris's seven agents, but adds a layer.
Watch out: The warnings page says try Hermes's built-in Kanban board first and add Paperclip only if it falls short.
Replaces: Granola. On-device dictation and meeting notes. The Next New Thing cites about 12,000 stars.
Watch out: Mac only, so Nuno Tavares (on Windows) could not test it.
Replaces: paid voice-to-text apps (the video names no single one). Free, works on several platforms, uses on-device models such as NVIDIA Parakeet. Keeps a short history, lets you add custom words, can tidy text with an AI model, and can unload the model to free memory.
For you: A free dictation option on Chris's Mac.
Watch out: The tidy-up step sends text to the cloud model you choose; plain dictation stays on device.
Replaces: ElevenLabs. Free, local voice cloning, voice design, dubbing, transcription, dictation and audiobook making in one app, using the OmniVoice model (600+ languages). Eric Michaud says setup took about 10 minutes and the clone is 'absolutely usable'. Claimed 61,000 stars.
For you: Ties to Chris's ElevenLabs voice setup for Hermes; could cut that cost.
Watch out: Jack Roberts says it is not as good as the best paid voices such as Fish Audio.
The channels behind Chris's 105 saved videos, ranked by how many of his videos each one made. Use it to see who teaches what, and in what style.
From your videos: All 105 videos from 53 channel labels (52 channels once Nate Herk's two labels are merged); 17 channels have 2 or more videos. Counts are from his saved set, not the channels' whole output. Plus 14 widely followed AI channels from the web that are not in his set (marked From the web), ranked by how relevant they are to agents, self-hosting, Claude and coding; each was checked as active in 2026.
11 videos in Chris's set. Clear, structured explainers with titles like 'X in N minutes': local and open-source models, Hermes and OpenClaw setups, and Obsidian as agent memory. She shows her own agent team (Coder, Lifebot, Taco Bot) and is careful about security: she avoids strangers' skills and runs audits twice a day. Two more videos in Chris's set are likely hers but the channel was not confirmed (see below).
8 videos in Chris's set. Practical Hermes workflows with copy-paste prompts: morning check-ins, memory wiki, client pulse, bot teams and token-saving habits. Calm, step-by-step style, with a business angle (50 ways to sell Hermes). His prompts are usually in the first description link.
7 videos in Chris's set. Weekly round-ups of trending GitHub repos and free, open-source swaps for paid tools, with two hosts. Fast and broad rather than deep; several episodes are Zapier-sponsored and names are often garbled in captions.
6 videos in Chris's set. Enthusiastic Hermes power user: use cases, lessons from months of daily use, new features, and comparisons with Dots, Grok Bot and Muse. Also pushes 'agentic income' from agent-run sites. He is also the guest on Greg Isenberg's Hermes Desktop video.
5 videos in Chris's set. Beginner-friendly explainers of every Hermes concept and level, plus repo round-ups (open-source Adobe-style apps, free Claude add-ons). States clear safety rules: no keys in chat, least access, no sending email.
4 videos in Chris's set. Long 'full course' walkthroughs and live tests of new models (GPT-6 Astra), Gemini Notebook and Hermes phone control. Often points to his paid Agent OS community.
4 videos in Chris's set. Weekly AI news round-ups: new models, image, video and voice releases, with benchmark comparisons and sponsored segments. Good for what is new, not for how-to.
4 videos in Chris's set. Calm, step-by-step beginner tutorials: a Hostinger VPS, server hardening, Dokploy deploys, and Hermes from install to a Kanban team. Closest to Chris's own hosting setup.
3 videos in Chris's set. Business-coach frameworks for agents and prompts (AGENT framework, Rule of R, MAPS) and a method to make AI write like you. Few named tools; promotes his own products.
3 videos in Chris's set. Long lists of Claude and Hermes skills, plus an honest money experiment where an agent's Etsy shop made $4.55 in 10 days.
3 videos in Chris's set. Obsidian as a working system and backend for agents: daily notes as a database, a human and machine vault split, and daily commands. Also covers open-source replacements for paid software.
2 videos in Chris's set. Interview format with guests: Alex Finn on Hermes Desktop, and Theo Taba on running an AI-native company with a markdown brain and skill chains.
2 videos in Chris's set. Tests many repos for Claude and says what to skip, plus a look at Kylon, a shared human-and-agent workspace (a sponsored request). Listed in the index under two names, 'Nate Herk' and 'Nate Herk | AI Automation', counted as one channel here.
2 videos in Chris's set. High-energy home-lab tutorials: running agents as an IT department in Paperclip, and monitoring every service and Hermes agent with Checkmk, with alerts that agents fix.
2 videos in Chris's set. Long build-and-sell courses: agent teams sharing subscriptions, and a three-hour Hermes course covering a paper trading bot, a health coach, Jarvis mode and selling setups to clients.
2 videos in Chris's set. Tours of Hermes desktop updates and skill picks, quoting GitHub star counts.
2 videos in Chris's set. Careful, detailed builds: a three-layer LLM wiki in Obsidian with Git save points, and Hermes bot mode with human approval and pruned tools.
Seven free GitHub repos for Claude Code, from Last 30 Days to GStack and Playwright MCP.
Principles and a mini build of an n8n research agent that writes PDF reports.
Six Hermes skill picks, including Planning with Files, RTK and Mantis security skills.
Covers the open-source Artcraft suite of Adobe-style apps: a hands-on test of each app, why a 'clean room' build is legal, and why MCP control beats screen clicking. Fair verdict: good for learning and personal work, not yet for paid client jobs. Includes a Higgsfield sponsor segment.
Turns Claude Code into a four-layer personal AI operating system (MAPS) with a short map file.
A $0 home-lab intranet over Tailscale to replace cloud subscriptions; no AI content.
Turns a Mac mini into a private local AI hub with Ollama and AnythingLLM.
Tests GPT-6 Astra building video games with Unreal Engine and Blender.
Ten handy command-line tools such as croc, VisiData and oxker.
Zapier-sponsored guide to connecting an agent to 9,000 apps by MCP, SDK or connectors.
Nine free agent skills, with Nvidia's Skill Inspector to scan them first.
Walks through his free Claude Code 'Jarvis' assistant with voice and Obsidian memory.
Runs uncensored local models offline in LM Studio.
Has Claude Opus 5.5 build a mobile app, website and launch video.
Nine new Claude Code mods, such as Cache Tax and Blast Radius.
Builds a first MCP server and client in Python with FastMCP.
A custom mission-control dashboard for 16 Hermes agents; the prompts are paid despite the title.
Deploys and manages a Hermes 'AI employee' from Claude Code on a Hetzner VPS.
Interview with a Google AI engineer on the fastest path into AI engineering.
Argues a RAG pipeline with a Claude Code hook beats an Obsidian second brain; package is paid.
Previews his own personal-OS harness, to be open-sourced once stable.
Eleven practical Hermes use cases, including a cost router across cheap models.
His always-on, provider-agnostic agent stack for running 100+ projects.
Checks eight viral GitHub repos and says which to skip.
Weekly AI news led by Claude Duet, Anthropic's always-on agent.
Builds a real customer-service bot with Hermes on a VPS and Telegram.
Eight scheduled ChatGPT Work agents that handle his recurring tasks.
Fifteen changes in Claude: short corrective skills, hooks, routines and cloud sessions.
Vendor demo adding a no-code AI agent to a Squarespace site.
Free open-source Mac apps that beat paid ones.
Twelve self-hosted apps to leave big tech, such as Jellyfin and Immich; no AI content.
Demos Claude Motion with Higgsfield and Remotion for code-made video.
Two videos where the presenter says she is Tina but the channel was not stated: a no-code first agent in n8n, and essential AI skills for 2026.
Builds an uncensored offline AI with Ollama and PrivateGPT; the channel name was not shown.
About 229K subscribers. Teaches building AI agents and using AI coding assistants, project by project, with a weekly upload; he has also covered an open standard for the Karpathy-style LLM Wiki. Listed as 'Very Active' in Oct 2026.
For you: Closest match to your work: agents, self-hosting and vault-style memory.
About 152K subscribers. 'Agentic engineering' with Claude Code: CLAUDE.md, hooks and MCP set up for real work, with a stated dislike of hype. Weekly uploads, 'Very Active' in Oct 2026.
For you: Goes deeper on the Claude Code hooks lesson in Masterclass.
Anthropic's own Claude channel: feature walkthroughs, 'How Anthropic uses Claude' playlists and Cowork how-tos. It posted 'Getting started with Claude Code Projects' about two weeks before 25 Oct 2026. Anthropic also runs a separate company channel, @anthropic-ai.
For you: Primary source: check here before trusting a third-party claim about Claude.
About 833K subscribers. Posts talks from the AI Engineer conferences by engineers, founders and researchers; growing about 13.7% a month in Oct 2026.
About 137K subscribers. Calm, technical walkthroughs of new language models and agent frameworks. Weekly uploads, 'Very Active' in Oct 2026.
About 566K subscribers. Long project courses on Claude Code, Codex and n8n, including a 4-hour Claude Code full course and a 3-hour advanced course from 2026.
Watch out: Many videos frame tools as ways to 'build and sell'; take the technique, skip the pitch.
About 421K subscribers. Project-based AI coding and agent builds. Weekly uploads, 'Very Active' in Oct 2026.
About 234K subscribers. Jason Zhou, a product designer, shares hands-on AI experiments and product builds. Uploads about every two weeks; marked 'Moderate' rather than 'Very Active'.
About 2.1M subscribers. Beginner-friendly coding tutorials; his 'FULL Claude Code Tutorial for Beginners in 2026' is a step-by-step start.
About 12M subscribers. A charity posting free full-length courses; 2026 uploads include a Claude Code for Beginners full course and a 3-hour 'AI Agents For Beginners' course.
AI news, model benchmarks and comparisons of Claude against other models, plus live news streams. His channel page showed an upload from two days before it was checked.
Watch out: News pace is fast; treat 'X just changed everything' titles as news, not advice.
About 573K subscribers. Daily videos on web development and TypeScript with strong opinions on AI coding tools; he also makes T3 Chat. 'Very Active' in Oct 2026.
About 4.3M subscribers. Very short, fast code tutorials and tech news, including quick first looks at AI coding tools. Weekly uploads, 'Very Active' in Oct 2026.
In-depth, carefully sourced coverage of new models, research papers and AI safety, rather than tutorials. Recent 2026 uploads include videos on AI lab security and model releases.
For you: Good for keeping up with models without hype; no setup steps.
Longer walk-throughs that follow one video from start to finish: what you will build, the steps in the video's order, and what you end up with.
From your videos: 14 tutorials, one per video, from 14 of the 105 videos. 10 were checked against full transcripts; 4 (marked) follow the notes only. Most of the 105 are news, tool lists or opinion, not step-by-step builds. Plus 7 free official tutorials from the web (marked From the web).
What you will build: a desktop-app Hermes with a provider, a self-made skill, cron jobs, four tiers of memory (files, session search, Honcho, an Obsidian vault) and a free local model.
For you: Her four memory tiers map well onto Chris's vault plus the planned Qdrant index for Cerebra.
Watch out: She runs a local Qwen 3.6 35B on a 64 GB Mac Studio; smaller machines need smaller models.
What you will build: a Hermes agent that runs all day on a rented server, thinks with a cheap OpenRouter model, talks to you on Telegram, and also opens on your Mac through the desktop app. About 30 minutes, no code.
For you: Chris already has Hermes under Coolify with Discord; the useful parts are the remote-gateway desktop step, key limits and the troubleshooting order.
Watch out: His test run cost 13 cents for 105 messages, and three quarters of the reading came from cache; your numbers will differ.
What you will build: a mental model of Hermes (surfaces, models, memory, skills, tools, cron, sub-agents, profiles) and a one-week order for setting it up without overload.
For you: Chris is past day 7; the 'seven mistakes' list is the useful check for his fleet.
Watch out: His seven mistakes: too many tools early, saving everything to memory, cheap models on hard work, every idea a cron, unchecked sub-agents, profiles before workflows, treating Hermes as magic.
What you will build: a Hermes 'first AI employee' with a home, a brain, a phone channel, a daily job and a skill it wrote itself, with approval gates for anything risky.
Watch out: He warns about prompt injection: give the agent its own machine and leave approval gates on.
What you will build: a three-layer vault (raw sources, wiki, schema) that a coding agent fills and checks for you, with Git save points, a firewall that keeps the agent in one vault, and an optional free local model.
For you: Brainiac's filing job; the allow-list idea can also keep agents out of the vault archive folder.
Watch out: He says get the cloud version working well before switching to a local model, which makes more mistakes.
What you will build: a Hostinger KVM 2 server with the Dokploy panel, a domain with SSL, an app deployed from GitHub, a Postgres database and Umami analytics.
For you: Chris uses Coolify for the same job; this is a comparison, not a switch.
Watch out: Steps are the video's sections in order, taken from the notes; the full transcript was not re-read, so exact clicks are not listed.
What you will build: a free Checkmk Community server that watches Linux hosts, network gear, Docker containers, Proxmox, websites, Windows and Hermes agents, with alerts to Slack and to an AI helpdesk.
For you: Sentinel's lane. On Chris's VPS, any new service goes through Coolify and Sentinel's review first.
Watch out: He says the free edition needs more manual work (plugins, piggyback folders) than the paid ones; community plugins are 'install at your own risk'.
What you will build: an n8n workflow that takes a topic from a form, writes five search queries, researches them through OpenRouter, and saves the result to an Airtable record.
For you: Chris already runs n8n with a research workflow for the agent team; the task-splitting and aggregation-database ideas transfer directly.
Watch out: His full 13 to 14 page version is in a paid community; only the mini build is shown.
What you will build: an n8n agent that researches a topic, turns it into audio and emails it, with a guardrail and simple evaluations.
For you: The evaluation step is the part Chris's research workflow may lack.
Watch out: Steps are the video's sections in order, taken from the notes; the full transcript was not re-read, so exact clicks are not listed.
What you will build: a flight-booking MCP server with resources, tools and prompts, connected to a coding assistant, plus a simple client.
For you: Useful background for Sentinel when reviewing any MCP server before it is added.
Watch out: He warns that remote and community servers need checks on authentication, privacy and trust.
What you will build: 'Amber', a support agent for a made-up dragon courier company, with business knowledge files, a public Telegram bot locked to owner-only admin commands, and a cheap daily report from a Python script.
For you: The soul file, knowledge folder and script-plus-cron ideas fit any of Chris's agents; his bots stay private.
Watch out: The allow-all variable makes the bot public; only use it with the admin lock in place.
What you will build: a morning command centre, a memory wiki, a content-research workflow, long /goal runs, a private mission-control dashboard, and a skill-building loop.
For you: Chris's planned agent dashboard should follow the same rule: gated sign-in and no secrets.
Watch out: His demo built a Telegram bot that reads client chats; that needs care with other people's messages.
A free self-paced course: 12 lectures, about 1.5 hours of video, one quiz and a certificate. It covers the agent loop, Plan Mode, the Explore, Plan, Code, Commit routine, context commands, CLAUDE.md, sub-agents, skills, MCP and hooks.
For you: Good for tightening Chris's CLAUDE.md and hooks; the quickstart page also points to Claude Academy.
Checked 26 Oct 2026
What you will do: install Claude Code, sign in, and use it on a real project, from asking questions to committing a change. Free to read; using it needs a Claude plan or Console account.
Checked 26 Oct 2026
A short walkthrough that deploys a ready-made nginx image, so you learn Coolify's project, resource, port, domain and deploy screens without any code.
For you: A safe practice run before changing anything in Chris's live Coolify projects.
Checked 26 Oct 2026
What you will do: install Hermes, choose a model provider, have a first chat, check sessions resume, then add a chat gateway, tools and skills.
Watch out: Skills from the hub still need Sentinel's review before install (see the warnings tab).
Checked 26 Oct 2026
A reading order in three levels: beginner (about 1 hour), intermediate (2 to 3 hours: messaging, memory, cron, skills) and advanced (4 to 6 hours: architecture, building tools and skills). It also has paths by goal, such as a Telegram or Discord bot.
For you: The intermediate level matches what Chris's seven agents already use.
Checked 26 Oct 2026
n8n's free hands-on courses, each with a certificate and badge: a quickstart that ends with a working AI agent, then three Foundations courses.
Watch out: The courses are free for everyone. A paid certification exam 'coming soon' could not be confirmed on the page on 26 Oct 2026.
Checked 26 Oct 2026
What you will do: create a tailnet (your private network), add two devices, then learn names (MagicDNS), sharing and access rules. The free Personal plan covers six users. Tailscale's own 10-minute video covers the same ground.
Checked 26 Oct 2026
What you will build: LM Studio on Windows, Linux or Mac running a small model with no internet.
Watch out: Steps are the video's sections in order, taken from the notes; the full transcript was not re-read, so exact clicks are not listed.
What you will build: an offline model running in Ollama, then PrivateGPT on WSL Ubuntu so you can chat with your own PDFs.
For you: Chris is on Mac, so the WSL step does not apply.
Watch out: Steps are the video's sections in order, taken from the notes; the full transcript was not re-read, so exact clicks are not listed. The video's focus is an 'uncensored' model.
What the videos say about 'uncensored' open models: models whose safety training has been loosened or stripped, run on your own machine so nothing is filtered and nothing leaves your computer. This tab stays factual and spells out the risks.
From your videos: Only two of the 105 videos are mainly about this: 'Private & UNCENSORED Local AI Model' and 'Build Your Own UNCENSORED AI Running Completely Offline'; a handful more touch local models in passing, and none explains how a model is made uncensored. Plus 10 entries from the web to fill gaps (marked From the web): abliteration and the research behind it, Dolphin and abliterated families, malware in model files, licences, Hugging Face rules and Canadian law.
Many abliterated models are published by individual uploaders, for example huihui-ai's abliterated Qwen builds. One such card (Huihui-Qwen3.8-27B-abliterated) says safety filtering 'has been significantly reduced', recommends research or controlled use only, not production or public-facing commercial apps, and says the uploader bears no responsibility.
Watch out: No benchmarks are given on that card. An uploader's disclaimer means the risk is yours.
Checked 26 Oct 2026
Dolphin 3.0 was curated and trained by Eric Hartford, Ben Gitter, BlouseJury and Cognitive Computations. Dolphin 3.0 is built on Meta's Llama 3.1 8B under the Llama 3.1 licence. Its card tells the operator 'You are the one who decides the guidelines', and that 'you are the creator and originator of any content you generate with Dolphin.'
Watch out: Older Dolphin builds on Ollama (dolphin-mixtral, about 26 GB for 8x7b) are Apache 2.0 but over a year old; the card's message is that all responsibility sits with you.
Checked 26 Oct 2026
Creating or possessing sexual images of children is illegal in Canada whether photographed or AI-generated. Bill C-16, the Protecting Victims Act, received Royal Assent on 18 June 2026: it extends the ban on sharing intimate images without consent to sexual deepfakes, makes threatening to share them an offence, and raises the maximum penalty to 10 years.
Watch out: Running a model offline changes nothing legally; the law is about what is made and shared. Information only, not legal advice.
Checked 26 Oct 2026
Hugging Face's content policy (effective 10 April 2025) bans non-consensual sexual content, any sexual content involving minors, malware, hate and harassment, and fraud. Uploaders must take steps to reduce the risk their models are used to make child sexual exploitation or terror material.
Watch out: A model being hosted there does not mean it was vetted.
Checked 26 Oct 2026
An uncensored model will produce content a hosted model blocks, which can include unsafe, unlawful or harmful material. Running one offline does not make the output safe to act on or to share, and you remain responsible for how it is used.
In Labonne's test, the abliterated Daredevil-8B scored lower than its source model on every benchmark he ran. A further training pass (DPO, a preference-tuning method) won back most of the loss, but not on GSM8K, a maths benchmark.
For you: Backs up the existing point that removing safety also removes a quality guard.
Checked 26 Oct 2026
In February 2024 JFrog found about 100 malicious models on Hugging Face. One PyTorch model used Python's pickle format so that simply loading it opened a reverse shell to an outside server. Hugging Face flags unsafe pickle files but, at the time, did not block downloads (it removed that model after reports).
For you: Exactly why Sentinel reviews before anything is installed.
Watch out: Prefer the safer safetensors format and named uploaders; never load a pickle file from an unknown account.
Checked 26 Oct 2026
Stripping refusals does not only unlock blocked topics; it can make a model more willing to give confident, wrong or harmful answers, because the same training that made it refuse also curbed bad output. Treat anything it says as unverified.
For you: Clashes with your 'always give me facts, check first' rule: an uncensored model is less trustworthy, not more.
Editing a model does not cancel its licence. Google's Gemma Terms cover 'Model Derivatives', including modified Gemma, and require passing the Prohibited Use Policy on to anyone you share it with. Gemma 4 has its own separate licence. Meta's Llama policy bars using Llama to break the law or risk bodily harm.
Watch out: Removing refusals does not remove your obligations under the licence.
Both videos find models by searching one word and name no author or repo. Downloading edited model weights from an unnamed source is the same trust problem as installing an unread skill, which your donts list warns against.
Watch out: Your rule is Sentinel reads anything before it is installed and you prefer named authors; neither video gives one.
If there is any case for a private local model in your setup it is privacy, not removing safety. A standard local model (plain Gemma 4 or Qwen through Ollama or LM Studio) gives the offline, private benefit without stripping the safety training.
For you: A safer middle path for Mantis's private lane than an uncensored build.
The offline video goes further and sets up PrivateGPT, a local tool that uses RAG (retrieval augmented generation, a way to give a model access to your own files) so you can ask questions over your PDFs and notes privately on localhost. It runs on WSL (Linux on Windows) with Python set up through pyenv and Poetry.
Watch out: This is the fiddly part of the video, with several Python steps; the presenter admits he had to fill gaps between two guides.
JarvesExplains uses LM Studio, a desktop app for Windows, Linux and Mac that downloads and runs local models. He searches the word 'uncensored' in its model list to find variants. He names no exact repo.
Watch out: If it shows 'failed to load model', he lowers the context length and GPU offload. Picking a model by searching one word is the opposite of vetting it.
Named in passing as uncensored variants: an uncensored Gemma 4 (about a 4 billion / 6 GB build, with a smaller 2B option), an uncensored Qwen 3.5 4B described as a 'thinking' model for lower-end PCs, and uncensored Llama builds. For weaker machines the Qwen and Llama options are suggested over the larger Gemma.
Watch out: The base models (Gemma, Qwen, Llama) are normal open models; it is the edited 'uncensored' copies, from unnamed sources, that carry the risk.
The offline-AI video uses Ollama to download and run a local model with 'ollama run'. The presenter again finds a variant by searching 'uncensored' in the Ollama library and does not name a specific model.
Watch out: Ollama is a mainstream local-model tool; the risk here is the particular model chosen, not Ollama itself.
Fine-tuning trains a model further on new examples to change its behaviour. Abliteration skips training and edits the weights once. Dolphin-style 'uncensored' models use the fine-tuning route instead: trained on data meant to make the model follow the operator's system prompt rather than the maker's rules.
Watch out: Either way, the result is a changed copy of someone else's model, made by a third party.
A normal chat model is trained to refuse some requests. An 'uncensored' model has had that refusal behaviour loosened or removed, so it answers things the standard version would block. JarvesExplains demonstrates this by asking the same question to a normal Gemma 4 (it refuses) and an uncensored Gemma 4 (it answers), both running locally. The videos also use the word 'abliterated' for this kind of edited model.
Watch out: The videos show the difference but never explain how the change is made, and name no safe, vetted source for these edited weights.
Arditi and colleagues (2024) found that in 13 open chat models, up to 72 billion parameters, refusal is controlled by a single direction inside the model. Erasing it stops refusals; adding it makes the model refuse even harmless requests. The authors say this shows the brittleness of current safety fine-tuning.
For you: Background only; useful for understanding why open-weight safety is easy to remove.
Checked 26 Oct 2026
Abliteration edits a model's weights directly, with no training. Researchers compare the model's internal activity on harmful and harmless prompts, find a 'refusal direction', and remove it, so the model stops refusing. Maxime Labonne's June 2024 Hugging Face post made the method widely known.
Watch out: The same post says the technique shows how fragile safety training is and raises ethical questions.
Checked 26 Oct 2026
Both videos stress privacy as the draw: the model runs on your own computer, works with the Wi-Fi off, and your data is not sent to any company. One presenter frames it as a fallback for when the internet is down.
For you: The privacy angle fits Mantis's walled-off private lane, but an uncensored model is a different and riskier thing from a private one.
Real jobs the presenters hand to their AI agents, from morning check-ins to home-lab alerts. Each one says how it works, what it uses and which of Chris's agents it would suit.
From your videos: 53 of the 105 videos feed this tab with a concrete use case; most are Hermes demos, and business ideas outnumber personal ones. Earnings claims are rare and small. Plus 7 personal-life entries from the web (marked From the web): family history, Canadian census, health privacy, Home Assistant, sim racing telemetry and learning. Money entries now note the Canadian rules (GST/HST, Etsy reporting, Stripe in Canada). Updated 26 Oct 2026: 2 entries from full transcripts of x2F3fdn82GU and BrBQRNIDRS8, plus a BAnQ Quebec records entry from the web.
Instead of adapting to someone else's app, ask for one that fits your problem. Example: Voxel Viewer, built with Claude to open a personal dental 3D scan that only paid software could read.
For you: Cyborg could do the same for sim-racing telemetry files.
Watch out: Copying a paid app's code or paywalled features crosses a line; the hosts of BrBQRNIDRS8 call that theft. Build from scratch for your own files.
Tina Huang writes a spec in Discord. Her Coder agent passes it to Claude Code to build, and a local Qwen model reports back. Hermes can also delegate to Codex or Claude Code through built-in skills.
For you: Cyborg/Vic, for code and iOS work.
Alex Finn has run /goal for over 24 hours. The trick is to ask the agent to write the goal prompt first, with details and limits. He then moves the code to Claude Code or Codex to finish.
For you: Cyborg/Vic; cap turns and spend first.
Watch out: Jack Roberts says a Hermes goal has a 20-turn budget by default; long runs cost tokens.
Jason Lee has Claude Opus 5.5 copy an invoice app feature for feature, design 22 screens with /design, build a landing page and make a 20-second launch video with the /brag skill.
For you: Cyborg/Vic for the iOS app; Forge for the video.
Sharbel A. adds a Telegram bot to his client chats. It reads the chats and gives him a daily list of clients who have gone quiet or seem unhappy. He built it by answering his morning priority question, and Hermes wrote most of it.
For you: Business only; Chris has no client chats today, so this is an idea to park.
Watch out: The bot reads every message in those chats, so the clients' privacy is at stake.
Alex Finn asks Hermes to open a competitor's site in its own browser, click around, read the console and write a full technical breakdown: stack, features, payments, analytics. He then hands the markdown file to a coding agent.
For you: Cerebra (research) writes it, Cyborg uses it for builds.
Python Simplified puts Hermes on a VPS, gives it a persona and five business documents, and opens it to the public on Telegram. Admin commands stay locked to one Telegram ID. A daily cron job runs a Python script to report on the day's chats.
For you: Pattern for any public-facing bot; Sentinel should review before anything is opened.
Watch out: One setting, TELEGRAM_ALLOW_ALL_USERS=true, opens the bot to everyone.
Several presenters let an agent read email and draft replies, but not send them. Sharbel A. keeps it draft-only. Moe Lueker triages Gmail. Dan Martell first builds a voice style guide from 20 recent emails so drafts sound like you.
For you: Alfred, with drafts only.
Watch out: Jack Roberts says not to let an agent send email at all.
Moe Lueker has a sub-agent find and enrich leads, with Apollo as one option. Lead Gen Jay connects Hermes to LinkedIn and Instagram DMs through Unipile and posts across platforms with Upload-Post. Sharbel A. lists lead generation as the first of 50 Hermes businesses.
For you: Not in Chris's current plans; listed so the option is known.
After each call, the agent pulls the transcript and turns action items into tasks. Sharbel A. uses Fireflies into Notion. Tina Huang's company bot reads Granola notes through its MCP. Jack Roberts connects Granola with OAuth.
For you: Alfred, filing the tasks; Brainiac can file the notes in the vault.
Greg Isenberg's guest runs one macro skill that fires smaller skills in order: a micro-site skill, a copy skill and a QA skill. Slack gets a ping when the proposal is ready. The same idea runs a five-step prototype-and-test chain.
For you: Forge could chain editing steps the same way.
Sharbel A. lists 50 services you could sell with Hermes, from lead generation and support triage to bookkeeping admin and workflow monitoring. Samin Yasar suggests a setup fee plus a retainer, and a 'risk reverser' so a first client is not locked in.
Watch out: These are ideas from the presenters; neither video shows client revenue. In Canada: register for GST/HST once taxable sales pass CAN$30,000 over four quarters (CRA).
Sharbel A. runs a webhook watcher (he calls it Sentinel) that watches sign-ups and site pages and alerts him on Telegram. Payments sit behind an approval gate.
For you: Sentinel (Chris's agent) is a natural owner, read-only.
Sharbel A.'s Nova skill researches video ideas and checks demand in VidIQ. Riley Brown's YouTube researcher skill reads transcripts to find podcast guests. Alex Finn's agent pulls a transcript and sets a cron quiz to help him learn it.
For you: Cerebra for research; Brainiac files the notes.
Sharbel A. builds a researcher bot and a creator bot under his main Hermes. They talk in a group room, draft an Instagram carousel, check each other, and only the main bot reports to him. Each bot gets only the tools its job needs.
For you: Matches Chris's seven-agent split; Forge does the editing.
Watch out: Cap turns in group rooms so bots do not loop and drain tokens.
Moe Lueker has an agent make a daily post with Nano Banana 2 or GPT Image 2, and uses Grok to read X for content gaps. Riley Brown drafts and schedules X posts through Typefully.
AI Samson asks Claude to rebuild one of his YouTube thumbnails inside PhotoCraft as a layered file, with each part named. He can then change the text or swap the background without starting over.
For you: Forge.
Watch out: Early software; check the result layer by layer.
Remotion and HyperFrames let an agent write an animation as code. Claude Motion does the same inside Claude, but only on Team and Enterprise plans per the videos. Tristen O'Brien adds Higgsfield as a custom connector for image and video models.
For you: Forge (video and editing).
Dan Martell has the AI interview you over four or more rounds to build a 'brain system' doc. A second prompt studies your recent emails, Slack and docs to write a voice style guide with banned words.
For you: Store it in the vault so Alfred and Forge share one voice.
Gran Turismo 7 sends live telemetry such as tire temperatures, RPM and position. Nenkai's open-source PDTools (MIT licence) first made it readable on a PC. The most-starred dashboard on GitHub's gt7 topic is snipem/gt7dashboard; smaller projects add lap comparison and AI driving coaches.
For you: Cyborg/Vic's sim-racing lane; complements Chris's existing GT7 HUD.
Watch out: Most AI-coach repos on the gt7 topic have under 10 stars; treat them as experiments.
NetworkChuck monitors his Hermes agents (cron jobs, gateway, model login) in Checkmk. When the gateway goes down, Checkmk opens a ticket in Paperclip, and an agent named Ron restarts it. Another agent added the rest of his hosts through the Checkmk API.
For you: Sentinel watches; fixes need Chris's yes under his rules.
Watch out: He gives the agent an admin automation account in Checkmk; that is broad access.
Home Assistant's Model Context Protocol Server lets an AI client such as Claude check states and control exposed devices. It signs in with OAuth, and needs an admin account by default.
For you: Alfred could add 'doors locked, heat set' to the evening brief.
Watch out: A cloud agent needs Home Assistant reachable from the internet; keep exposed devices to a short list.
Alex Finn puts Tailscale on every device so Hermes can reach them all. From his phone he asks it to fetch a file off his desktop, install a local model on another Mac, or test a site on a different machine.
For you: Cyborg/Vic; links the VPS, Mac, iPhone and iPad privately.
Watch out: It gives the agent reach into every device, so scope it tightly.
Better Creating runs Ollama and AnythingLLM on a Mac mini and reaches it from an iPad with Screens 5. Other videos run uncensored or private models with PrivateGPT or LM Studio, fully offline.
For you: Mantis could use a local model for private wellbeing notes.
Watch out: The uncensored-model videos say little on safety or reliability.
Tina Huang runs OpenClaw's security audit twice a day and sends Hermes security reports to an alerts channel. AI LABS shows Mantis Plan and Mantis Reflect (security review skills by Google) run on cron.
For you: Sentinel's regular job. Note the skill set shares a name with Chris's Mantis agent.
The Next New Thing and Paul J Lipsky show a Claude Code script, Clean Your TV, that removes ads from a TCL TV.
Watch out: It changes a device's software; no owner or repo was named in the videos.
Home Assistant's Ollama integration lets you ask about and control exposed devices through a model on your own computer, with no cloud model. Only models that support tools can control devices.
Watch out: Home Assistant calls device control experimental; smaller models make more mistakes.
Nimrod presses the iPhone Action button, records a voice note, and Whisper turns it into a Hermes task. Sharbel A. and Pav Rusovs send Telegram voice notes to a home machine.
For you: Alfred, for reminders on the go; Discord voice notes would be Chris's version.
Dubibubi let an agent on GPT-6 Astra run a digital-product Etsy shop for 10 days, using eRank for keywords. It made 4.55 USD. The useful lesson was a loop that cut tokens per listing from 4% to 0.33% of his weekly limit.
Watch out: A clear example of how small agent earnings can be. In Canada: Etsy reports Canadian sellers to the CRA after 30 physical-goods sales or CAN$2,800 in a year, or any made-to-order digital sale.
Alex Finn's bot picked an idea, built a directory site, and maintains it with routines. It sends a daily newsletter through Resend and takes sponsorships through Stripe. He says it has made 1,100 USD so far, all from two sponsorships.
Watch out: He says traffic came from his large X audience; the bot asks approval before payments or sign-ups. In Canada: Stripe serves Canadian businesses, and once taxable sales pass CAN$30,000 over four quarters you must register for GST/HST (CRA).
Tina Huang has a Hermes cron job move Apple Notes into Obsidian, alongside a daily brief with audio and a nightly health check.
For you: Brainiac (vault filing).
Library and Archives Canada has online census databases from 1825 to 1926, including 1851 to 1921 nationally and early Lower Canada (Quebec) years. Ancestry.ca needs a subscription, but LAC notes many public libraries offer it free. These are the primary records to check AI hints against.
For you: Useful for Montreal and Quebec branches of the family tree.
A scheduled brief of calendar, email, news and notes. Tina Huang adds an audio version through NotebookLM, and another of her workflows writes the digest straight to Apple Notes. Pav Rusovs runs a morning digest routine.
For you: Alfred (briefings).
FamilySearch uses AI and handwriting recognition to make old handwritten records searchable (Full-Text Search), and its AI Research Assistant suggests ancestors to review. An agent can then file each find with its source link. FamilySearch itself warns that AI makes mistakes.
For you: Fits Chris's family-history project: Cerebra searches, Brainiac files, and each fact keeps its source.
Watch out: FamilySearch says to always check the original image, even when an AI transcript exists.
Tina Huang's Lifebot reads her Pomodoro log and to-do list in Obsidian and her Apple Health steps. Samin Yasar gave Hermes his blood test PDF and sleep data; it set daily cron reminders for workouts and supplements, which he answers by voice note.
For you: Mantis, kept private and walled off.
Watch out: Samin's supplement list came from his agent reading his blood work; no clinician check is shown.
Apple Health records are encrypted on the iPhone, and with two-factor authentication they are end-to-end encrypted in iCloud, so Apple cannot read them. Each app you allow chooses what it can see. Sending that data to a cloud AI takes it outside that protection.
For you: Mantis's lane, kept walled off; a local model avoids sending health data to OpenRouter.
Watch out: Anthropic says AI answers still need checking for high-stakes decisions; health is one.
Alex Finn has Hermes build a site of every topic and daily log, like an automated diary that also feeds its memory. Tina Huang and Wanderloots run the Karpathy LLM wiki pattern in Obsidian: ingest, query and lint.
For you: Brainiac; matches Chris's agent diaries.
Watch out: Eric Michaud keeps agents off the human side of the vault.
Every morning at 9:00 the agent asks your number one priority, suggests three ways it can help, and starts what you approve. Alex Finn says it also teaches the agent about you over time.
For you: Alfred (reminders and briefings).
BAnQ (Quebec's national library and archives) keeps Quebec's baptism, marriage and burial registers from 1621 up to 100 years ago; newer ones stay with Quebec's civil registrar. Many are scanned on BAnQ numérique: pick 'Registres de l'état civil', choose a parish, then a year, and read the index and register pages in the viewer. BAnQ's search guide describes this with no fee or account mentioned; the fees BAnQ lists are for certified copies ($55 for Quebec residents, $350 for others).
For you: For the George Cooper research: Cerebra searches, Agent 99 fact-checks, and every fact keeps its BAnQ link. Pages are in French.
Watch out: Records newer than about 100 years are not at BAnQ, and some registers are not scanned yet. Certified copies currently have long delays.
Metics Media sets scripted cron jobs for a receipt filing clerk and a Barcelona-to-London flight price watch.
For you: Alfred for the watch, Brainiac for filing.
Alex Finn, on Greg Isenberg's show, runs a cron job that scans Reddit and X for AI business opportunities and builds a dashboard and prototypes.
For you: Cerebra (web research).
AI Foundations builds an n8n agent that refines queries, splits the work, researches with Perplexity and stores sections and citations in Airtable before making a PDF. A separate no-code n8n build turns research into audio and emails it.
For you: Cerebra, using Chris's existing n8n.
Wanderloots runs an orchestrator, a researcher and a librarian bot. The librarian files finished reports into an Obsidian LLM wiki and proposes changes for his approval. Only the orchestrator writes to external memory.
For you: Cerebra researches, Brainiac files.
Julian Goldie SEO walks through Gemini Notebook (the renamed NotebookLM): add sources, chat with citations, and make audio overviews, slides, flashcards and quizzes.
When learning a new topic, give the model your own sources and tell it to answer only from them, quote the exact lines, and say 'I don't know' when the sources are silent. Anthropic lists these as the main ways to cut made-up answers.
For you: Works with Cerebra for research and the notebook study flow already on this tab.
Vibe coding means building software by describing what you want to an AI. This tab covers the tools the videos name, the habits that make it work, and the mistakes presenters warn about.
From your videos: 38 of the 105 videos feed this tab, but most only name tools in passing. The habits come mainly from Nate Herk, Simon Scrapes, Sharbel A., Jason Lee and Julian Goldie. 12 entries from the web add the official Anthropic and OpenAI guides, the vibe coding definition and security research on AI-built apps (marked From the web).
Anthropic names five common mistakes: the kitchen-sink session that mixes unrelated tasks, correcting over and over, a CLAUDE.md so long that rules get lost, trusting code that looks right but misses edge cases, and open-ended 'investigate' requests that read hundreds of files. The fixes: /clear, a fresh prompt, prune CLAUDE.md, always verify, and scope the search or use subagents.
Watch out: Anthropic's own words: 'If you can't verify it, don't ship it.'
Checked 26 Oct 2026
Sharbel A. calls 'Build me an app, make no mistake' a bad prompt: it leaves too much undefined. Big jobs need a designed goal with success criteria and a way to check the result.
Many titles promise a full app 'in one prompt'. In Brendan Jowett's test the model quietly used three Mario clone repos from GitHub. Dubibubi let an agent run an Etsy shop for 10 days and it made 4.55 US dollars.
Sharbel A. says switching model, effort, fast mode, or MCP and plugin toggles in the middle of a session resets the prompt cache, so the next turns cost far more. Compacting does the same.
A skill is instructions the agent obeys. The Next New Thing reported a science skill that pushed users to a paid product from inside its own file. Nuno Tavares says to run a security scan on any repo link first, and Futurepedia shows Nvidia's Skill Inspector for that.
For you: Already Chris's rule: nothing installs without Sentinel's review and his yes.
Nate Herk tells newcomers not to install all his picks on day one: if something breaks, you will not know which one caused it. Pick the one that fixes your current problem.
Mason Fitzgerald calls CLAUDE.md a soft instruction that is not enforced. Simon Scrapes points to hooks instead: checks that run every time, for example when Claude says it has finished. His /goal loop also needs visible proof in the chat before it stops.
Nate Herk cites a JetBrains test: Caveman cut about 8.5% of output tokens against 65% advertised, RTK saved nothing, and Ponytail saved about 10%, half its claim. He also now skips Claude Mem, which sends usage data by default, and Context Mode, which has a usage bug.
Jay E of RoboNuggets shows Claude Code mods. Blast Radius pauses risky deletes and lists the files at stake, Replay Theater steps through each edit Claude just made, and File Tree shows touched files and which are committed. Container Use runs the agent in a sandbox (a sealed box) so it cannot break your machine.
For you: A delete guard suits Chris's archive-never-delete rule.
Watch out: Mods need Claude Code 2.1.287 or newer and are temporary until saved as a plugin.
gstack's /codex command asks Codex to review what Claude wrote. The Next New Thing also lists an open-source Alibaba code reviewer with a VS Code extension.
For you: Sentinel already plays reviewer for Chris; a second model is the same idea for code.
Julian Goldie says that without examples Codex makes bland pages that look 'like a ChatGPT website straight out of the box'. Jason Lee says a good reference matters more than the model. Anthropic's frontend-design skill bans generic fonts and forces a design direction.
For you: Chris wants mock-ups in Claude Design before building, which is this habit.
Agents often code against old versions of a library. Context7 feeds current library documentation into the chat, and JetBrains' Go modern guidelines do the same for the Go language.
For you: The Swift version of this is the SwiftUI agent skill in the 'Building apps' tab.
Pav Rusovs keeps CLAUDE.md to about 40 lines that point to other files, and measured a task at 30 seconds with the map against 62 without. Simon Scrapes suggests nested folders with pointers so context loads only when needed. Claude Code now also reads AGENTS.md, a shared instruction file.
Matt Pocock's Grill Me skill makes Claude ask about 20 questions before it acts. Sharbel A. 'meta prompts' a big job: he asks the agent to help design the goal prompt and to ask clarifying questions first. Dan Martell calls this reverse prompting.
For you: Suits Chris's one-question-at-a-time style; the agent can ask one at a time.
Nate Herk warns that Claude will say 'this works' without having clicked the buttons. Microsoft's Playwright CLI opens a real browser, clicks through the app like a customer and writes tests you can re-run. Because it is a command-line tool, not an MCP, it does not dump whole pages into Claude's context, so it uses fewer tokens.
Method packs are sets of skills that make the agent work like a careful engineer. Superpowers enforces brainstorm, plan, test and review; gstack gives role skills from Y Combinator's Garry Tan; Matt Pocock's pack includes a test-first (TDD) skill. Nate Herk says Addy Osmani's is easiest for newcomers.
For you: Superpowers has a Hermes plugin, so it could suit Cyborg after Sentinel's review.
Watch out: ECC alone holds 292 skills; giant packs cost tokens on every turn.
Addy Osmani's agent-skills run /spec to write a real spec with you, /plan to break it into small tasks, and /build to work through them one at a time with a check on each. Nate Herk says the spec also raises edge cases you would miss, such as double bookings or time zones. Because the plan lives in files, a fresh session can pick up where the last stopped.
For you: Plan files could live in the Obsidian vault so Brainiac files them.
Watch out: Simon Scrapes says the Claude Code team is considering removing plan mode, so do not rely on it alone.
Jason Lee feels Claude gets worse past about 300,000 tokens of chat. He asks for a handoff prompt: Claude writes the project's files, colours and instructions into a handoff file, then gives a short prompt to paste into a new chat in the same folder. Sharbel A. adds /rename and /resume to save a session and /rewind to undo turns.
For you: Matches Chris's rule of one file per log, newest first, so nothing gets buried.
Git keeps a history of every change, so a bad edit can be undone. Alex Finn connects GitHub so his bot pushes code and pull requests (proposed changes), and Jack Roberts backs up his whole agent to GitHub daily. A Git log replay tool even plays the history back as an animation.
For you: Chris's vault already syncs to GitHub; the same habit fits every Cyborg project.
Sharbel A. explains that the whole chat is re-sent every turn. He uses /context to see what fills the window, /usage and /cost for spend, and turns off unused MCP servers. Codeburn shows where tokens go, and Simon Scrapes's /advisor lets a cheap model ask a stronger one only when stuck.
For you: Fits the OpenRouter budget limits Chris already runs per agent.
Anthropic suggests four phases for bigger changes: read the code in plan mode, write a plan, build it, then commit. For a large feature, have Claude interview you and write SPEC.md, then build it in a fresh session. Skip the plan when you could describe the change in one sentence.
Checked 26 Oct 2026
Anthropic's top tip: Claude stops when work looks done, so give it a test, a build or a screenshot to compare that says pass or fail. Then it can fix and re-check alone. Ask for evidence, such as the test output, not just 'it works'.
For you: Matches Chris's rule to check things before reporting them.
Checked 26 Oct 2026
A fresh session or subagent (a helper agent with its own memory) reviews better because it did not write the code. Anthropic shows a Writer and Reviewer pattern and an adversarial review step where a subagent checks the diff. Tell the reviewer to flag only real correctness gaps, because chasing every finding leads to over-engineering.
For you: The same idea as Sentinel reviewing Cyborg's work.
Checked 26 Oct 2026
OpenAI says each prompt should give a goal, context, constraints and a 'done when' line. Keep a short, accurate AGENTS.md with run, test and lint commands. Keep one chat per coherent unit of work, and use git worktrees when parallel threads touch the same files.
Watch out: OpenAI lists scheduling a recurring task before it is reliable by hand as a common mistake.
Checked 26 Oct 2026
If you have corrected Claude twice on the same thing, the chat is full of failed attempts. Anthropic says to clear it and start again with a better prompt that uses what you learned. A clean session with a better prompt almost always does better.
Watch out: Checkpoints only undo Claude's own file edits, not changes made by commands. They do not replace git.
Checked 26 Oct 2026
Both Anthropic and OpenAI say to start with tight approval and sandbox settings and loosen them only when needed. Claude Code lets you pre-approve safe commands, run commands in a sandbox (a walled-off area), or use auto mode, where a separate model blocks risky actions. OpenAI lists giving full machine access too early as a common mistake.
For you: Fits the house rule that nothing is installed without Sentinel's review and Chris's yes.
Checked 26 Oct 2026
Andrej Karpathy coined 'vibe coding' in a post on X on 2 Feb 2025, meaning accepting AI changes without reading them, for throwaway weekend projects. Simon Willison says it is only vibe coding when you do not review the code; he will not commit code he could not explain to someone else. He calls the careful, accountable version 'vibe engineering', which leans on tests, plans, docs, git and code review.
Checked 26 Oct 2026
Veracode tested 100+ models in 2025 and across 80 coding tasks found security flaws in 45% of cases; models failed to block cross-site scripting (a common web attack) in 86% of cases. Its 2026 report says things barely moved: an average security pass rate of 56%, even though the code nearly always runs.
For you: A clear case for Sentinel reviewing every build Cyborg makes.
Checked 26 Oct 2026
Researchers found 19.7% of packages suggested by code models did not exist; when 500 such prompts were rerun 10 times, 43% of the fake names came back every time. Attackers can register those names with malware inside. Open-source models made this mistake far more often (21.7% on average) than commercial ones (5.2%).
Checked 26 Oct 2026
CVE-2025-48757 (a public security record) covers sites built with Lovable up to 15 Apr 2025. Weak database rules let strangers read or change data without logging in. Lovable disputes the record and says each customer must protect their own data.
Watch out: A prompt-to-app builder does not make the database safe for you.
Checked 26 Oct 2026
Simon Willison warns about agents that have three things at once: your private data, contact with untrusted content, and a way to send data out. An attacker can then trick the agent into leaking your data. He says the only safe fix is to avoid that combination, not to trust prompts or guardrail products.
For you: Worth checking for each Hermes agent: does any one of them hold all three?
Watch out: He says a guard that stops '95% of attacks' is a failing grade in web security.
Checked 26 Oct 2026
Anthropic's coding agent, run in the terminal or the desktop app. It is named in 39 of the videos, more than any other tool. Samin Yasar uses it for planning work, and Pav Rusovs runs it on both his laptop and a VPS (a rented server).
For you: A natural home for Cyborg's code lane. Hermes also ships built-in skills to hand jobs to Claude Code or Codex when asked.
OpenAI's coding agent, with a desktop app, plugins and cloud routines (scheduled jobs that run with the app closed). Samin Yasar says it is best for development tasks, and AI Edge uses it for a second-opinion code review. Julian Goldie and Brendan Jowett build sites, apps and games in it with GPT-6 Astra.
For you: Samin Yasar also uses Codex as the 'doctor' that repairs Hermes, a useful backup role.
Watch out: Julian Goldie's claim that GPT-6 Astra beats Claude is his opinion, in a course tied to his paid community.
A code editor with AI built in. corbin uses it for his Zapier demo and says Claude Code and Codex work the same way. In Alex Finn's video, Cursor cloud agents inside Grok Bot built the first version of a live site in about two minutes. A Cursor plugins repo connects it to 90+ apps.
An open-source coding agent for the terminal that works with models from many providers. The Next New Thing cites about 220,000 GitHub stars. Nimrod runs it beside Claude Code and Codex on one always-on PC.
For you: Because it works with other models, it could use Chris's OpenRouter models. Sentinel reviews first.
Antigravity, Windsurf, GitHub Copilot, Roo Code and the Warp terminal are named as coding tools, mostly in lists without a demo. OpenClaude is a fork of Claude Code that runs other models.
Watch out: The Next New Thing says OpenClaude's legality is unsettled.
Lovable and Bolt build a web app from a plain description in the browser. Bolt also connects to tools like GitHub and Notion through MCP. Tina Huang lists both among the coding tools on her AI roadmap.
Watch out: Both appear mainly as sponsors in these videos, not as tested picks.
Orca and Claude Squad run many agent sessions at once, each on its own task. Git worktrees (separate working copies of one project) stop them from editing the same files, and Worktrunk is a command-line tool that manages them for Claude Code.
Watch out: More agents means more spend; start with one.
Nimrod, a CTO running over 100 projects, uses T3 Code as a control panel over Claude Code, Codex and OpenCode. It has a mobile app, so he can start and check coding threads from his phone. Its T3 Connect feature links a few devices without Tailscale.
For you: Could let Chris watch Cyborg's coding threads from his iPhone.