OpenKB
An open-source command-line tool that compiles raw documents into a structured, interlinked wiki-style knowledge base using large language models, with reasoning-based retrieval for long documents and no vector database required.
OpenKB (Open LLM Knowledge Base) compiles raw documents into a structured, cross-referenced wiki using LLMs, powered by PageIndex's vectorless, reasoning-based retrieval for long documents, following an approach described by Andrej Karpathy in which an LLM writes summaries, concept pages and cross-references that are kept in sync automatically. Rather than rediscovering knowledge from scratch on every query as traditional RAG does, OpenKB compiles it once into a persistent wiki that stays current.
It supports a broad range of input formats, including PDF, Word, Markdown, PowerPoint, HTML, Excel, CSV and plain text, and understands figures, tables and images as well as text. Wiki pages follow the Google Open Knowledge Format specification, the resulting wiki is plain markdown compatible with Obsidian's graph view, and a bundled web UI lets a user browse the knowledge base, upload and compile documents, and stream queries and chats in the browser.