Skip to content

Repository files navigation

AI Second Brain

A practical skill that walks you through building a living, searchable, AI-powered knowledge base from your ChatGPT/Claude history, research, and daily inputs — then talking to it from your laptop or phone.

Maintained by Yo-Da Lai · Gold Coast, Australia
Works with Claude Code and Hermes Agent.


Why this exists

On 3 April 2026, Andrej Karpathy published a gist on building a personal wiki with a folder of markdown + an LLM. The idea spread fast for a reason: RAG re-discovers knowledge every time. A wiki compounds it.

This skill packages that pattern — plus an AI-history organiser process popularised by Alex Freedman and slash-command workflows shared by Greg Isenberg — into one guided setup.

Original packaging was released by Charlie Hills (MIT). This is Yo-Da Lai’s maintained fork: clearer installs, Hermes + Claude Code paths, a sturdier wiki scaffold, and a share page you can send people to.

Share page (send people here): https://ezauto399.github.io/ai-second-brain/
GitHub repo: https://github.com/EzAuto399/ai-second-brain
Personal site / work with me: https://yodalai.xyz


Install

Option A — Claude Code (recommended for most people)

git clone https://github.com/EzAuto399/ai-second-brain.git ~/.claude/skills/ai-second-brain

If that folder already exists, back it up first:

mv ~/.claude/skills/ai-second-brain ~/.claude/skills/ai-second-brain.backup

Open Claude Code anywhere and say:

Set up my AI second brain.

Update later:

cd ~/.claude/skills/ai-second-brain && git pull

Option B — Hermes Agent

# Personal profile example — adjust the skills path if your Hermes profile differs
git clone https://github.com/EzAuto399/ai-second-brain.git \
  ~/.hermes/profiles/personal/skills/research/ai-second-brain

Then in Hermes:

Load the ai-second-brain skill and set up my AI second brain.

If Hermes already has the built-in llm-wiki skill, this skill still helps for the full guided path (exports → vault → channels → slash commands). Use both: this skill for setup, llm-wiki for ongoing wiki ops.

Verify (Claude Code)

/skills

You should see ai-second-brain. If not, the clone path is wrong — re-run Option A.


What it does

flowchart LR
    A[ChatGPT / Claude<br/>exports] --> B[Step 1<br/>AI Brain]
    C[PDFs, notes,<br/>transcripts] --> D[Step 2<br/>Living Wiki]
    E[Gmail, calendar,<br/>meetings] --> F[Step 3<br/>Heartbeat]
    B --> G[Tagged, linked<br/>Obsidian vault]
    D --> G
    F --> G
    G --> H[Talk to it from<br/>laptop or phone]
Loading

Three stages — do all three, or pick one:

  1. AI Brain — turn ChatGPT + Claude history into tagged, linked markdown in Obsidian.
  2. Living Wikiraw/ sources in, structured wiki pages out (Karpathy-style, with a durable schema).
  3. Heartbeat — optional connectors + messaging channels + slash commands (/today, /ideas, /create).

Every step the agent can do, it does. Every step only you can do (exports, macOS permissions, browser logins) pauses until you confirm.


What you need


Trigger phrases

The skill should activate on things like:

  • “Set up my AI second brain”
  • “Build the Karpathy wiki”
  • “Organise my ChatGPT conversations in Obsidian”
  • “I exported my ChatGPT data, what now?”
  • “Help me set up iMessage / Telegram channels for my wiki”
  • “Create the /today, /ideas and /create slash commands”

Already done part of it? Say: “skip step 1, I’ve already done it”.


FAQ

Does this send my data anywhere?
No. Everything lives in folders on your machine. The only network calls are the ones your agent already makes to its model provider. Your conversations, wiki, and raw sources stay local.

Will it read my email automatically?
Only when you ask, and only for the threads relevant to the request. You control connectors and can revoke access anytime.

Claude Code or Hermes?
Both work. Claude Code is the path most newsletter-style walkthroughs assume. Hermes is great if you already run a multi-skill local agent (and may already include llm-wiki).

Can I uninstall the skill?
Yes:

rm -rf ~/.claude/skills/ai-second-brain
# or Hermes path:
rm -rf ~/.hermes/profiles/personal/skills/research/ai-second-brain

That removes the skill only. Your Obsidian vault / wiki data stays.

Does it work on Windows?
Steps 1–2 yes. Native iMessage channels are Mac-only — use Telegram/Discord/gateway options instead.

What if I don’t want messaging channels?
Skip that sub-step. You still get a living wiki on your laptop.

Business / team version?
A personal second brain is the right start. If you want a governed company brain + AI staff for a real business (approvals, channels, memory), that’s what I build for clients at yodalai.xyz (ClawConnect by Yoda Lai — AI staff platform).


Credits

Underlying ideas are not mine. This skill exists so setup takes minutes, not a weekend.

  • Andrej Karpathyoriginal gist for the living markdown wiki pattern (raw/, wiki compile loop, LLM as maintainer).
  • Alex Freedman (@alex2learn) — AI Brain export-and-organise process adapted in Step 1.
  • Greg Isenberg — slash-command-on-vault workflow that Step 3c is based on.
  • Charlie Hills — original MIT packaging that this maintained fork builds on.

I (Yo-Da Lai) maintain this fork: dual-agent install, sturdier wiki scaffold, channel options, and the public share page.


About

Yo-Da Lai — forward-deployed AI engineer on the Gold Coast, Australia.
I help non-technical business owners automate real work with AI systems that stay under human approval.

If this skill helped, send someone the repo link — or the Pages URL once it’s live on this repository.


Licence

MIT. Fork it, modify it, rebrand it — keep the credits to Karpathy, Alex, Greg, and the original packaging notice.

About

Build a living AI second brain from ChatGPT/Claude history + research. Claude Code + Hermes skill by Yo-Da Lai.

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors