llm-wiki

This repo turns Andrej Karpathy's "LLM Wiki" gist into a concrete local starter kit for an agent-maintained markdown knowledge base.

• raw/ stores immutable source material.
• wiki/ stores the compiled, synthesized markdown knowledge base.
• outputs/ stores saved briefs, reports, and durable answer artifacts.
• AGENTS.md, CLAUDE.md, and PROMPTS.md make the workflow portable across coding agents.
• python3 -m llm_wiki ... handles ingest, batch processing, indexing, search, linting, repair, and report generation.

Architecture

The core loop is: collect raw sources, process and synthesize them into a markdown wiki, answer questions against it, and file durable outputs back into the wiki.

Quick Start

Obsidian is optional. Any text editor plus a local-file-aware coding agent will work. The workflow lives in AGENTS.md, CLAUDE.md, and PROMPTS.md.

How To Use It

Treat this repo as a persistent thinking layer: a structured second brain that compounds knowledge over time.