RAGFlow
A retrieval-augmented generation engine that combines document parsing, chunking and retrieval with agent templates, deployable as a self-hosted Docker service or used through a hosted cloud version.
RAGFlow is a retrieval-augmented generation engine that combines a RAG pipeline with agent capabilities. It parses unstructured documents such as Word, Slides, Excel, PDFs, scanned copies, images and web pages using deep document understanding, splits them into chunks using configurable templates, and visualizes the chunking so a person can check and adjust it before it is used for retrieval.
Retrieval combines multiple recall methods with re-ranking, and answers are grounded with traceable citations back to the source chunks to reduce hallucination. It supports configurable LLMs and embedding models, a Python and JavaScript code executor for agents, agentic workflows with MCP, memory for agents, and data synchronization from sources such as Confluence, Notion, Google Drive and S3.
It is self-hosted with Docker Compose, requiring at least 4 CPU cores, 16GB of RAM and 50GB of disk, or used through a hosted cloud service. It is written in Go and Apache-2.0 licensed.