Ragpi

Ragpi fills a niche for teams that want an AI-powered documentation Q&A without sending data to third-party services. Its agentic RAG approach is smart: it doesn't just retrieve chunks but reasons about which documents to fetch based on the query. This gives more relevant answers than naive vector search alone. The hybrid search with RRF is a nice touch, combining semantic and keyword matching for better coverage. On the integration side, Discord and Slack support make it easy to drop into existing workflows. The web widget is handy for public docs. But here's the catch: it's entirely self-hosted. You need to manage a server, a vector database (likely Qdrant or similar), and an LLM provider key. There's no cloud version, so scalability and uptime are on you. Compared to something like Danswer (now Onyx) or QnA3, Ragpi is lighter but less feature-rich — no file upload, no analytics, no permission system. Where it bites: if your documentation is poorly structured or spans many formats, the connector options are limited to sitemaps, GitHub Issues, and READMEs. No generic web crawling or PDF support yet. Also, the project is relatively new; community support and documentation are thin. For teams that need a quick, managed solution, this isn't it. But for an open-source shop with DevOps chops, Ragpi offers a transparent, customizable AI assistant that keeps your data in-house. We'd reach for this when we want to experiment with RAG over documentation without vendor lock-in.