OpenKB

Why we included this project

OpenKB works differently from the usual retrieval pipeline: instead of pulling fragments from raw files on every question, it uses an LLM to compile your documents once into a persistent wiki of concept and entity pages, and keeps that wiki current as you add material. The design follows Andrej Karpathy's idea of compounding knowledge, so summaries and cross-references accumulate over time, and contradictions get flagged as they surface, rather than everything being re-derived on each query. Teams dealing with long or complex documents, such as research groups, legal or compliance units, and product documentation owners, get reasoning-based retrieval with no vector database to run, and native support for formats like PDF, Word, Markdown, PowerPoint, and HTML. The CLI exposes both the wiki-building layer and query, chat, and skill generators, so you can maintain the knowledge base directly or wrap it into an assistant. If you are tired of re-paying retrieval costs for every prompt, this is a genuinely different way to keep an internal corpus useful.