# LLM wiki

> **TL;DR.** LLM wiki is a pattern where an AI agent incrementally compiles a folder of plain-markdown entity pages from your raw sources, then answers questions against that structured wiki instead of re-retrieving the originals every time.

- **Category:** AI / Knowledge Management / Agent Workflows
- **Stage:** validating
- **Age:** 73 days
- **Origin date:** 2026-04-04
- **First detected:** 2026-04-21
- **Canonical URL:** https://earlyterms.com/term/llm-wiki
- **Sources:** 8 primary URLs

## Definition

LLM wiki is a pattern where an AI agent incrementally compiles a folder of plain-markdown entity pages from your raw sources, then answers questions against that structured wiki instead of re-retrieving the originals every time. Knowledge compounds in the filesystem rather than evaporating after each chat.

The term crystallized on [April 4, 2026](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f) when Andrej Karpathy posted a gist framing his personal research workflow — raw/ on one side, a wiki/ of cross-linked markdown on the other. It hit Hacker News at 296 points / 95 comments the same day, and within a week spawned a dozen Show HN clones and a domain rush across every major TLD.

## Example

Karpathy's own wiki reportedly grew to ~100 articles and 400k words while still navigable by a coding agent. Open clones followed within days: lucasastorian/llmwiki (570 stars), nashsu/llm_wiki desktop app (2k stars), and sdyckjq-lab/llm-wiki-skill — a Claude Code skill that ports the pattern into a reusable agent harness.

## Analogy

Like a personal Wikipedia that writes and edits itself while you read.

## Why it's emerging now

Karpathy's April 4 gist reframed RAG as a compiler problem — ingest once, cross-link forever — and agentic coding tools like Claude Code finally made the maintenance step cheap enough to automate. The HN thread crossed 95 comments in a day, and 12/13 TLDs on llmwiki.* were grabbed within two weeks.

## Related terms

- *parent:* context-engineering
- *competitor:* RAG
- *competitor:* GraphRAG
- *related:* agent-loop
- *related:* claude-code
- *related:* agents-md
- *related:* Obsidian
- *competitor:* NotebookLM
- *competitor:* DeepWiki
- *related:* knowledge graph
- *related:* second brain
- *alias:* idea file

## Sources

1. [Karpathy — llm-wiki.md gist](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f)
2. [Hacker News flagship thread (296 pts)](https://news.ycombinator.com/item?id=47640875)
3. [Analytics Vidhya — LLM Wiki Revolution](https://www.analyticsvidhya.com/blog/2026/04/llm-wiki-by-andrej-karpathy/)
4. [Data Science Dojo — tutorial](https://datasciencedojo.com/blog/llm-wiki-tutorial/)
5. [Global Advisors — Term briefing](https://globaladvisors.biz/2026/04/06/term-llm-wiki-andrej-karpathy/)
6. [MindStudio — build with Claude Code](https://www.mindstudio.ai/blog/andrej-karpathy-llm-wiki-knowledge-base-claude-code)
7. [Level Up Coding — Beyond RAG](https://levelup.gitconnected.com/beyond-rag-how-andrej-karpathys-llm-wiki-pattern-builds-knowledge-that-actually-compounds-31a08528665e)
8. [lucasastorian/llmwiki (reference OSS impl)](https://github.com/lucasastorian/llmwiki)

---
_Generated by EarlyTerms · https://earlyterms.com/term/llm-wiki_