llm-wiki-second-brain

A clean open-source template for building a Karpathy-style LLM Wiki second brain with Markdown, Git, and coding agents such as Codex or Claude Code.

What This Is

This template is for people who want a simple, durable knowledge system without building a full app.

The model is intentionally small:

• raw/ contains curated immutable source material.
• wiki/ contains LLM-maintained synthesized Markdown.
• AGENTS.md gives persistent instructions to the coding agent.
• the human adds sources and asks questions.
• the agent ingests, queries, lints, and maintains the wiki.

Topic pages should emerge from real source material. Do not pre-create many topic pages.

Why This Pattern Works

• Markdown stays readable without any special software.
• Git gives you review, history, and rollback.
• Obsidian works well as a browser and reader.
• Coding agents can maintain the wiki layer consistently over time.
• The system compounds: good query answers can be filed back into the wiki.

Repository Layout

Initial wiki/ contents should only be:

Do not create domain-specific topic pages initially. Let the agent create them during ingest when the sources justify them.

How To Use It

• Drop a curated source into the correct vault's raw/.
• Ask your coding agent to ingest it.
• Review the git diff.
• Commit when the changes look correct.