RAGFlow

RAGFlow is an open-source Retrieval-Augmented Generation engine that combines RAG with agent capabilities to form a context layer for large language models. It offers a streamlined RAG workflow adaptable to enterprises of any scale, turning complex, unstructured data into production-ready AI systems. It is built for developers who need a converged context engine with pre-built agent templates.

A team deploys RAGFlow with Docker Compose or builds it from source, then opens its Web UI to add an LLM, embedding model and reranker with an API key. RAGFlow ingests documents through deep document understanding and template-based chunking, then answers questions through agentic retrieval that can break a question down, gather supporting facts, and verify them across multiple rounds.

RAGFlow's Knowledge Compilation turns documents and datasets into Wikis, Graphs, Trees, Mind Maps, Timelines and Skills, and its agentic retrieval offers Low, Medium, High and Ultra thinking modes to match question complexity. It supports Word, PowerPoint, Excel, PDF, scanned and web-page data sources, configurable LLMs and embedding models, multiple recall with fused re-ranking, and visualized, traceable citations.

RAGFlow is open source under the Apache-2.0 license and can be tried through a hosted cloud service or deployed locally with Docker, recommended at 4 CPU cores, 16 GB RAM and 50 GB of disk space. Building the Docker image from source does not currently support macOS, though Docker deployment itself runs on Windows, macOS or Linux hosts.

Use cases

• Turning complex enterprise documents into production-ready AI systems
• Answering complex, multi-step questions by breaking them down and verifying evidence