RAGFlow

RAGFlow is an open-source RAG engine that starts at the unglamorous end of retrieval — parsing the documents — and builds up from there into hybrid search, an ingestion pipeline, and visual agent workflows that tie retrieval, tools and MCP integrations together. Its own framing is that the context layer is the thing to get right, and it names financial services and legal work as the settings it is built for.

Starting from parsing is the more honest architecture, because most retrieval failures are decided before retrieval happens: a table split mid-row, a scanned page treated as one blob, a heading that belonged to the section below it. No amount of vector-search tuning recovers from chunks that were wrong on the way in, which is why the parsing and chunking stage deserves as much attention as the model, and it is where most RAG pipelines silently lose quality. The trade is involvement: a self-hosted engine is yours to run and upgrade, and the visual workflow layer is a convenience over code you may prefer to keep in code.