Key Takeaways
- Apache 2.0 license β free to self-host, modify, and redistribute
- Built by InfiniFlow, an open-source RAG engine that fuses retrieval with agent capabilities
- Deep document understanding: layout-aware parsing extracts tables, figures, and structure rather than flattening a file to plain text
- Grounded citations: every answer links back to the specific chunk it was generated from, with a visual preview
- Self-hosts via Docker; minimum hardware is 4 CPU cores, 16 GB RAM, 50 GB disk
- Best fit: document-heavy workflows (contracts, financial statements, scanned reports); a lighter tool suffices for plain text-file Q&A
- A paid managed cloud (cloud.ragflow.io) also exists for teams that prefer not to self-host β this review covers the self-hosted, open-source deployment
π In One Sentence
RAGFlow is an open-source (Apache 2.0), self-hosted RAG engine that specializes in deep document understanding β layout-aware parsing of complex PDFs, tables, and scanned files β with answers that trace back to their exact source chunk.
π¬ In Plain Terms
Instead of treating a PDF as one long string of text, RAGFlow reads its layout β tables stay tables, footnotes stay attached to their page β and shows you which exact snippet of the document it used to answer your question.
What Is RAGFlow?
RAGFlow (github.com/infiniflow/ragflow) is an open-source RAG engine built by InfiniFlow, released under the Apache 2.0 license. Its own documentation describes it as fusing retrieval-augmented generation with agent capabilities to create a context layer for LLMs β in practice, a self-hosted application with a web interface for building document Q&A systems and agentic workflows.
- Deep document understanding: a layout-aware parsing pipeline (DeepDoc) extracts text, tables, and figures from complex file formats instead of treating every file as flat text
- Template-based, intelligent chunking that RAGFlow describes as "explainable" β you can see and adjust how a document was split before it is indexed
- Grounded citations with source-chunk visualization, so an answer can be checked against the exact passage it came from
- Agentic capabilities: a visual workflow builder for chaining retrieval, tools, and multi-step reasoning, plus Model Context Protocol (MCP) support
- Supports Word, Slides, Excel, TXT, images, scanned copies, structured data, web pages, and more as source documents
- Deployable via Docker for self-hosting, or through InfiniFlow's paid managed cloud (cloud.ragflow.io) if you prefer not to run infrastructure yourself
Deep Document Understanding & Citations β RAGFlow's Core Differentiator
Simple RAG is easy when a document is a clean text file. Business documents rarely are β a contract can contain nested tables, footnotes, headers, scanned pages, and cross-references, and a naive PDF-to-text conversion collapses all of it into an unstructured blob. RAGFlow is specifically built to parse that structure instead of discarding it.
πNote: RAGFlow also supports GraphRAG-style knowledge graph construction and configurable agentic reasoning depth on top of the core document pipeline β evaluate these against your own use case rather than assuming you need them for basic document Q&A.
How Does RAGFlow Differ From Quivr, Dify, and LlamaIndex?
All four tools touch retrieval-augmented generation, but they solve different problems. RAGFlow should be evaluated against ready-to-use document Q&A applications like Quivr, not against a vector database or a code framework β they compete for the same "point it at my documents and ask questions" buying decision.
How Do You Self-Host RAGFlow?
RAGFlow deploys via Docker. These are the documented steps from the project's own quickstart guide, current as of this review β always check the repository's docs for the latest release tag before deploying.
- 1Confirm your machine meets the minimums: CPU with 4+ cores (x86), 16 GB+ RAM, 50 GB+ free disk, Docker 24.0.0+, and Docker Compose v2.26.1+.
- 2On Linux, raise the kernel's memory map limit for the bundled search component: sudo sysctl -w vm.max_map_count=262144, then add vm.max_map_count=262144 to /etc/sysctl.conf so it survives a reboot.
- 3Clone the repository and check out a stable release tag: git clone https://github.com/infiniflow/ragflow.git, then cd ragflow/docker && git checkout -f v0.27.1 (use whatever tag is current when you deploy).
- 4Start the stack: docker compose -f docker-compose.yml up -d. Watch the logs with docker logs -f docker-ragflow-cpu-1 and wait for the service to report a successful start before continuing.
- 5Open http://<your-server-ip> in a browser (RAGFlow listens on port 80 by default), go to Model providers, and add the API key or endpoint for the LLM and embedding model you plan to use.
- 6Create a dataset, choose an embedding model and chunking method, upload your files, and click the play button to parse them β then review the resulting chunks and create a chat linked to that dataset to start asking questions.
Can RAGFlow run without an internet connection?
RAGFlow itself runs fully offline once deployed, but it needs an LLM to generate answers. Point it at a local model server (such as an Ollama-compatible endpoint) rather than a cloud API to keep the whole pipeline offline β check RAGFlow's model-provider settings for which local endpoints it supports.
How much disk space does RAGFlow actually need?
The documented minimum is 50 GB free, on top of the space your indexed documents and their generated chunks/embeddings will consume. Budget more for large document collections, since RAGFlow stores parsed chunks and embeddings alongside the source files.
Who Should Use RAGFlow?
RAGFlow's document parsing depth is a real advantage for messy files and a real cost β in setup complexity and hardware β for teams that don't need it.
RAGFlow vs. Alternatives
A side-by-side view of where each tool sits, using the categories buyers actually compare on: interface, document handling, and license.
Tool | Interface | Document parsing | Citations | License |
|---|---|---|---|---|
| RAGFlow | Web app + visual agent builder | Layout-aware / tables & scans | Traceable to source chunk | Apache 2.0 |
| Quivr | Web app | General-purpose text extraction | Source references | Open source |
| Dify | Web app + workflow builder | General-purpose (RAG is one module) | Source references | Open source |
| LlamaIndex | Code framework, no built-in UI | Configurable via loaders you choose | You implement it | MIT |
Common Mistakes When Evaluating RAGFlow
These mistakes come from treating RAGFlow as a drop-in replacement for a simpler tool, or the reverse β under-provisioning it and blaming the software.
Frequently Asked Questions
What is RAGFlow?
RAGFlow is an open-source (Apache 2.0) RAG engine built by InfiniFlow, specialized in deep document understanding. A layout-aware parser extracts tables, figures, and structure from complex documents, and every generated answer links back to the exact source chunk it came from.
Is RAGFlow free?
The self-hosted, open-source deployment is free under the Apache 2.0 license. InfiniFlow also offers a separate paid managed cloud (cloud.ragflow.io) for teams that prefer not to run their own infrastructure β that is a distinct product from the free self-hosted software this review covers.
What license is RAGFlow released under?
Apache License 2.0, which permits free use, modification, and redistribution, including in commercial products.
How does RAGFlow's citation feature work?
Every answer RAGFlow generates includes references to the specific chunks it was built from. The interface lets you preview the cited passage inside the original document, so you can verify the answer against its actual source rather than trusting the summary alone.
What makes RAGFlow different from a typical RAG tool?
Most RAG tools convert a document to plain text before chunking, which loses table structure, footnotes, and layout. RAGFlow parses layout first β tables stay tables β which matters specifically for complex business documents like scanned contracts and financial statements.
What document formats does RAGFlow support?
Word, Slides, Excel, TXT, images, scanned copies, structured data, and web pages, per the project's own documentation. Check the current GitHub repository for the complete, up-to-date format list before deploying, since RAGFlow ships new format support frequently.
Can I self-host RAGFlow?
Yes. RAGFlow deploys via Docker and Docker Compose on your own infrastructure. The documented minimum hardware is 4 CPU cores, 16 GB RAM, and 50 GB free disk.
What are the hardware requirements to self-host RAGFlow?
CPU with 4 or more cores (x86), 16 GB or more RAM, 50 GB or more free disk space, Docker 24.0.0 or newer, and Docker Compose v2.26.1 or newer, per RAGFlow's quickstart documentation.
RAGFlow vs. LlamaIndex β which should I use?
Choose LlamaIndex if you need a fully custom retrieval pipeline built in code, with your own choice of vector database and no packaged UI. Choose RAGFlow if a ready-to-use web application with strong document parsing already covers your requirements, without writing a retrieval pipeline yourself.
Does RAGFlow support AI agents, not just RAG?
Yes. Beyond document retrieval, RAGFlow includes a visual workflow builder for chaining retrieval steps, tools, and multi-step reasoning, along with Model Context Protocol (MCP) support for connecting external tools to an agent.