# llm-wiki

> A design pattern for building LLM-maintained personal knowledge bases in markdown

An architectural idea file by Andrej Karpathy proposing a persistent, compounding markdown wiki maintained by an LLM agent rather than transient RAG queries. Users feed raw documents into a repository while the AI organizes, cross-references, and updates topic pages viewed alongside an editor like Obsidian. Designed as a prompt spec, it is intended to be handed to a coding agent to build a customized local setup.

Website: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f

## What it is for

- Instruct a coding agent to scaffold an automated personal knowledge base
- Synthesize research across dozens of articles into cross-referenced topic pages
- Maintain living documentation from project notes, transcripts, and reading materials
- Audit note collections for contradictions and orphaned concepts via agent lint passes

## Features

- Separates architecture into immutable raw sources, agent-maintained wiki files, and an agent instruction schema
- Incremental ingest workflow that updates topic overviews, entity pages, and central catalogs automatically
- Append-only log tracking and index catalog to simplify retrieval without complex vector databases
- Compounding query responses that file deep answers and comparisons directly back into the wiki
- Automated linting passes to identify dead links, missing topics, and factual contradictions
- Integrates natively with plain markdown tools and Obsidian features like graph view and Dataview

## Pricing

Free public GitHub Gist

## Details

- Type: GitHub repo
- Category: [Productivity](https://xiry.app/topics/productivity.md)
- Topic: [Knowledge Management](https://xiry.app/topics/productivity/knowledge-management.md)
- Open source: yes
- Platforms: Web, other
- For: Developers, researchers, and knowledge workers using AI coding agents and markdown editors
- Alternative to: NotebookLM, Mem.ai
- Tags: pkm, obsidian, llm agents, markdown, knowledge base, rag

## Why it was saved

- [@karpathy, 4 April 2026](https://x.com/karpathy/status/2040470801506541998): A shareable spec of the idea instead of code — the agent customizes and builds it for the user's specific needs.

## Compare

- [Alternatives to NotebookLM](https://xiry.app/alternatives/notebooklm.md)

## Related

- [Notion](https://xiry.app/resources/notion): Connected workspace for notes, docs, databases, and AI automation. (Web app, Freemium)
- [Andrej Karpathy's LLM Knowledge Base Workflow](https://xiry.app/resources/andrej-karpathy-s-llm-knowledge-base-workflow): A workflow for using LLMs to compile and maintain personal wikis (Framework, Free)
- [Are.na](https://xiry.app/resources/are-na): Ad-free platform for saving, organizing, and connecting ideas into channels (Web app, Freemium)
- [Ars Umbris](https://xiry.app/resources/ars-umbris): Local-first IDE and framework for building typed, agentic knowledge graphs (Web app, Open source)
- [LLM Living Knowledge Base Pipeline](https://xiry.app/resources/llm-living-knowledge-base-pipeline): Architecture framework for building an LLM-maintained markdown wiki that compounds knowledge (Framework, Free)
- [YapThread](https://xiry.app/resources/yapthread): Turn Twitter bookmarks into a searchable AI knowledge base with citations (Web app)

Updated 10 October 2026. Page: https://xiry.app/resources/llm-wiki

Make your own library: Xiry takes your saved reels, tweets, Reddit bookmarks and Watch Later videos, makes them searchable and pulls the useful stuff out, like this library. It's invite-only for now; join the waitlist at https://xiry.app/#waitlist.