llm-wiki

A self-maintaining second brain for your codebase.

Your AI agent reads it, writes to it, and routes every question through it — so knowledge accumulates instead of evaporating between sessions. Based on the Karpathy LLM Wiki pattern, extended for software development workflows.

Works with GitHub Copilot, Claude Code, Codex, and any agent that reads markdown.

Get Started

New project — use this repo as a GitHub template

Click "Use this template" at the top of this page, or:

Then run local setup inside your new repo:

Existing project — install into your repo

Run this from your project root:

The script copies the wiki skeleton, .agent/ planning system, agent prompts, and scripts into your repo — without overwriting any existing files. Then run your agent's /wiki-onboard command to fill in project-specific knowledge.

Manual install (without curl)

• Download and unzip this repo: llm-wiki-main.zip
• Copy wiki/, .agent/, scripts/, .github/prompts/, .github/agents/, CLAUDE.md, AGENTS.md, CONVENTIONS.md, and PROGRESS.md into your project root.
• Add .agent/context.md to your .gitignore.
• Run bash scripts/onboard.sh to bootstrap your local agent context.
• Run /wiki-onboard in your agent.

What's Included

The template ships with a starter wiki built around software development — ready to use immediately and extend as your project grows.