LLM Wiki

Most retrieval setups answer each question from raw documents all over again. LLM Wiki instead has a language model read what you import and compile it into a structured, interlinked wiki that it keeps updating as sources change, an approach based on Andrej Karpathy's LLM Wiki pattern.

It ingests PDFs, Office files, EPUB and MOBI books, Org mode notes, images, web clips and batches of URLs, and each generated page traces back to its source. Links between pages feed a knowledge graph with community detection, vector search through LanceDB is optional, and a Chrome clipper, a local HTTP API and an MCP server let other tools work with the wiki. Models can come from OpenAI, Anthropic, Google, Ollama or a custom endpoint. The app is built with Tauri for macOS, Windows and Linux and released under GPL-3.0.