infiniflow/ragflow

RAGFlow is an open-source Retrieval-Augmented Generation engine. The idea behind Retrieval-Augmented Generation, or RAG, is that a large language model is given relevant pieces of your own documents at the moment of answering, so its responses are grounded in your data instead of just its training. RAGFlow combines this with what it calls Agent capabilities, giving developers a context layer they can put between their data and an LLM to build production AI systems. The engine is designed around what the README calls "Quality in, quality out". It performs deep document understanding to extract knowledge from unstructured data in many formats, Word documents, slides, Excel files, plain text, images, scanned copies, structured data, web pages, and more. Long documents are split using template-based chunking, described as intelligent and explainable. Answers come with grounded citations: chunked text can be visualized for human intervention, and key references are viewable so users can trace where each answer came from, framed as a way to reduce hallucinations. Around this core, RAGFlow offers an automated RAG workflow with configurable LLMs and embedding models, multiple recall paired with fused re-ranking, and APIs for integration. The README lists support for agentic workflows, MCP, and data synchronization from Confluence, S3, Notion, Discord, and Google Drive. Someone would use RAGFlow to build a question-answering or assistant product on top of their own documents. It is self-hostable via Docker, with stated requirements of at least 4 CPU cores, 16 GB RAM, and 50 GB disk. The project is in Python and licensed under Apache-2.0.