infiniflow/ragflow

Open-source RAG engine that goes past the standard chunk-and-retrieve pattern by folding in agent capabilities for multi-step retrieval. Where most RAG setups treat the retrieval step as a single lookup, RAGFlow can chain retrievals and tool calls to build a richer context before hitting the LLM. 80k stars suggests real traction, not just hype. The self-hosted angle matters here: if your documents contain anything a legal or compliance team would object to sending through a managed API, this is the obvious shortlist entry. Reservation: the marketing copy oversells it, and the actual differentiator versus a well-tuned LangChain or LlamaIndex pipeline is narrower than the README implies. Worth benchmarking against your specific document corpus before committing. -> Best for: SaaS team of 2-5 building a document-grounded product that cannot use hosted RAG services for compliance reasons