RAGFlow

RAGFlow helps teams build AI agents that answer questions using their own documents. It extracts content from complex files and retrieves supporting material that readers can check.

• Document extraction handles Word files, slides, spreadsheets, images, scans, structured data and web pages, including layouts with tables.
• Search and citations combine vector search, BM25 keyword matching, custom scoring and reranking with traceable references for answers.
• Multi-step retrieval lets an agent break down complex questions when needed, gather more context over multiple rounds and verify evidence.
• Knowledge organization lets you inspect and adjust document chunks or compile content into wikis, graphs, mind maps and timelines.
• Visual workflows combine retrieval and tools with optional MCP connections, using configurable language, embedding and reranking models.

RAGFlow is open source under Apache 2.0. You can self-host it through Docker on Windows, macOS or Linux, or use its separate hosted cloud service. Model connections can use local services or external providers.

Document layout analysis, OCR and table recognition run on the CPU. The recommended starting configuration is four CPU cores, 16 GB RAM and 50 GB of available storage. Requirements depend on document processing, data volume and concurrency; local models and optional components may require additional resources.