Build Your LLM WikiNo Code. No Maintenance.

An LLM wiki is a personal knowledge base that an AI model compiles for you: you save raw sources, the model writes summaries and cross-linked concept pages, and you query that layer over time. The term comes from Andrej Karpathy, and Recall is one product that builds an LLM wiki automatically. Unlike one-off file uploads in a chat (RAG), an LLM wiki compounds as you keep adding to it.

Install the Recall browser extension, save sources in one click, browse your library and graph, chat across your knowledge base, save good answers as notes, and review tags or graph gaps occasionally. Recall handles the compile step automatically on every save, so you skip scripts, Obsidian plugins, and hand-maintained markdown folders.

Recall is the closest off-the-shelf product to the LLM wiki Andrej Karpathy described when he said there was room for a real product instead of a hacky collection of scripts. It runs the same shape out of the box: ingest, compile, browse, query, and maintain a compounding wiki with no code. You trade some DIY control, like a local git vault, for a tool that works immediately.

No, they are different. RAG (retrieval-augmented generation) means attaching documents to a chat session and asking a question now, while an LLM wiki compiles sources into persistent summaries and links so later questions get richer over weeks. Recall implements that compiled layer as saved cards and a knowledge graph, so chat behaves more like querying a wiki than re-uploading files each session.

No, not with Recall. Andrej Karpathy uses Obsidian as a reader while the LLM writes the wiki, but Recall combines reading, saving, linking, and chat in one product. You can bulk import an existing Obsidian vault if you already have one, then use Recall for new saves and library-wide chat.

Recall runs the same five-part LLM wiki workflow as a DIY Obsidian-and-Claude stack, with far less upkeep. A DIY stack gives you local files and custom formats like Marp or matplotlib, but ingest takes about 10 minutes per source and monthly health checks often run 30 to 60 minutes. Recall keeps saves under a minute each, with optional 10 to 15 minute monthly maintenance.

Yes. Recall lets you save any chat answer back into your library as a note card, where it gets the same summaries, tags, and links as an ingested source. That mirrors the output step Andrej Karpathy described, where useful answers become part of the wiki for the next question.

Yes, Recall is free to start, with unlimited saves and notes. The free plan includes a limited number of AI summaries per month, or you can use the Recall MCP to connect your library to another AI subscription. Recall Plus unlocks unlimited AI summaries and chat across your library.

With Recall, you can ask cross-source synthesis questions, request grounded answers with citations, or run lint-style prompts like what you saved about a topic but have not connected yet. Recall chat can draw on your compiled library, the web, or both, using the AI model of your choice on supported plans.

Recall ingests articles, YouTube videos, podcasts, PDFs, newsletters, TikToks, and personal notes into your LLM wiki. You can save from the browser extension, share to the mobile app, paste URLs in the web app, or bulk import Markdown from an Obsidian vault.