llm_wiki

LLM Wiki is a cross-platform desktop application that reads your documents and automatically builds a structured, interlinked knowledge base from them. Rather than the common approach where an AI searches your documents and answers questions from scratch each time, this tool works differently: it processes your documents once, generates organized wiki pages from them, and keeps those pages updated as you add more material. The result is a persistent personal knowledge base that grows and improves over time.

The application is based on a methodology by Andrej Karpathy for building personal wikis with language models. Documents you add are stored as raw sources, and the LLM generates wiki pages in a two-step process: first analyzing the content to identify key entities, concepts, and connections to existing wiki entries, then writing the actual pages. Each page includes links back to the source documents it came from, and an incremental cache skips re-processing files whose content has not changed.

The desktop interface uses a three-column layout: a file and knowledge tree on the left, a chat panel in the center, and a preview panel on the right. There are dedicated views for search, a knowledge graph, a lint checker, a review queue for items flagged for human judgment, and a deep research mode that runs web searches and pulls results back into the wiki automatically.

The generated wiki files are compatible with Obsidian, a popular note-taking application that uses the same wiki-link syntax, so the folder can be opened directly in Obsidian without any conversion.

Other features include extracting and captioning images from PDFs, optional vector-based semantic search, a Chrome extension for clipping web pages into the knowledge base, and automatic detection of new or changed files in your source folder.