RAGFlow

RAGFlow is a leading open-source retrieval-augmented generation engine with 92,000+ stars on GitHub that fuses deep document understanding with agent capabilities to build a superior context layer for LLMs.

Overview

Most RAG systems fail at the ingestion step: messy PDFs, scanned forms, tables and layouts get flattened into bad chunks, and no retriever can recover from that. RAGFlow attacks the problem at the source with a deep document-understanding pipeline that parses complex formats into structured, semantically rich representations, then layers hybrid search — vector similarity, BM25 and custom scoring with re-ranking — on top. More recent releases grew it into a full platform: an ingestion ETL pipeline, visual agent orchestration that combines RAG, tools and MCP servers, and ready-made workflow templates for domains like investment research, legal analysis and manufacturing support. The project is Apache-2.0 licensed, maintained by InfiniFlow, and updated continuously, with repository activity in October 2026.

Key Features

• Deep document parsing – understands layout, tables and mixed formats instead of naive text extraction.
• Hybrid retrieval – vector search plus BM25, custom scoring and re-ranking for high-precision answers.
• Agent orchestration – build agents visually, combining RAG, tools and MCPs in one workflow.
• Ingestion pipeline – built-in ETL cleanses and structures multi-format data for retrieval.
• Grounded answers – citations and traceable context reduce hallucination in production Q&A.
• Self-host or cloud – deploy the open-source engine yourself or use the hosted service.

💳 Pricing

RAGFlow is free and open source under the Apache License 2.0 for self-hosting. The hosted cloud offers a free tier, then Starter at $29/month and Pro at $129/month, with Enterprise plans for BYOC and on-premises deployment.

Who Is It For

Teams building serious knowledge assistants and document Q&A — especially in document-heavy fields like finance, legal and manufacturing, where parse quality decides whether RAG works at all.