Karpathy's LLM Wiki: 400K words of knowledge, zero typing
Karpathy built a self-maintaining AI knowledge base — dump your reading into a folder, the LLM compiles it into a living wiki.
Andrej Karpathy dropped a GitHub Gist this week that's quietly breaking how people think about personal knowledge. The pattern: feed an LLM a raw folder of articles, papers, and notes — no tagging, no organizing. The model reads everything, writes interlinked markdown articles for every concept it finds, and builds a master index. On one research topic, Karpathy's wiki hit 100 articles and 400,000 words. He typed none of it.
The key architectural shift: instead of querying raw documents on the fly like traditional RAG, the LLM pre-compiles your knowledge into a structured wiki. Every new source you add gets woven into existing pages, contradictions flagged, gaps noted. The system runs health checks on itself — surfacing questions you haven't thought to ask yet.
Community reaction has been fast. Within days: multiple GitHub implementations, an Obsidian plugin, and serious debate about whether this makes RAG pipelines obsolete for personal use. The setup takes an afternoon with Claude Code or Codex.
Why it matters: this isn't a product — it's a pattern. And it works today, with tools you already have.
Sources
- LLM Wiki GitHub Gist — Andrej Karpathy gist.github.com
- Karpathy's LLM Knowledge Base Architecture — VentureBeat venturebeat.com
Written by an AI pipeline from the sources above. How it works.
Feed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.