RAGFlow
ragflow.io
- Free plan
- Open source
- macOS
- Linux
- Web
- API
- Self-hosted
What RAGFlow is
RAGFlow is an open-source RAG engine and agent platform for building a context layer for AI agents. Its ingestion pipeline processes different kinds of data into semantic representations for retrieval, while search brings together vector search, BM25, custom scoring, and re-ranking. Visual workflows connect RAG, tools, and MCPs for agent orchestration. Teams can connect online, local, or OpenAI-compatible model providers for knowledge bases, chats, search, and agents. Documented connectors include Notion, Google Drive, S3, Microsoft Teams, REST API, and Sitemap; HTTP and Python APIs are also available, and a chat assistant can be embedded in another webpage. RAGFlow offers cloud and self-hosted deployment. Its Docker documentation sets a minimum build requirement of 4 CPU cores, 16 GB RAM, and 50 GB disk. The free plan includes five apps, one team member, 0.1 GB of dataset storage, and 500 credits per month, but no API key. The Starter plan is $29 a month and includes an API key.
Who it suits
RAGFlow suits teams building retrieval-backed AI agents, knowledge bases, or chat experiences, including the financial services, legal and compliance, manufacturing, and education sectors listed by the company. Self-hosting requires meeting its documented minimum build requirements.
Good to know, and what to watch
- Combines vector search, BM25, scoring, and re-ranking.
- Visual workflows connect RAG, tools, and MCPs.
- Offers HTTP and Python APIs.
- Free plan includes 500 credits per month.
- Free plan does not include an API key.
- Self-hosted Docker build requires 4 CPU cores and 16 GB RAM.
- RAGFlow does not maintain ARM Docker images.
Plans
Every plan and priceEnterprise
Free
Starter
Pro
Free plan
The verdict
RAGFlow combines data ingestion, retrieval, and agent workflows with both cloud and self-hosted deployment options. Check plan limits and deployment requirements, especially if you need an API key or ARM-based Docker image.
What the maker publishes
- Agent orchestration
- Visual workflows combine retrieval, tools and MCP services for agent construction. source
- Agent workflows
- Its visual workflows integrate RAG, tools, and MCPs for agent orchestration. source
- API and embedding
- RAGFlow provides HTTP and Python APIs and supports embedding a Chat assistant in a third-party webpage with an iframe. source
- API migration
- RAGFlow 1.0 moves its public REST backend from Python to Go and documents endpoint and parameter changes. source
- Architecture limitation
- The official Go image targets Linux AMD64; ARM64 requires a matching dependency image and native libraries. source
- Chat channels
- Supported chat-channel integrations include Discord, DingTalk, Telegram, WeCom and Feishu/Lark. source
- Chunking control
- Users can inspect chunking results and intervene in parsing through configurable chunk templates. source
- Company founded
- LinkedIn lists InfiniFlow as founded in 2023. source
- Company headquarters
- LinkedIn lists InfiniFlow headquarters in Shanghai, China. source
- Company legal name
- Crunchbase lists the legal name as InfiniFlow Inc. source
- Company size
- LinkedIn lists InfiniFlow as having 11–50 employees. source
- Data connectors
- Documented data sources include Notion, Google Drive, S3, Microsoft Teams, REST API, and Sitemap. source
- Data processing
- Its built-in ingestion pipeline cleanses and processes multi-format data into semantic representations for retrieval. source
- Deep document understanding
- RAGFlow supports parsing complex documents such as scanned PDFs, tables, images and slide decks. source
- Deployment complexity
- Independent analysis says Docker deployment also runs Elasticsearch or Infinity, MinIO, MySQL and Redis, increasing operational complexity. source
- Developer interfaces
- Independent documentation describes web UI, REST, Python and JavaScript APIs, MCP and agent workflows. source
- Document ingestion
- Its built-in ingestion pipeline cleans and processes multi-format data into semantic representations for retrieval. source
- Enterprise support
- The Enterprise plan lists dedicated support and a custom SLA. source
- External tools
- Agents can connect to external MCP servers over Streamable HTTP or SSE; stdio servers require a gateway or another service exposing a supported endpoint. source
- Hardware limitation
- An independent review warns that deployments with less than 16 GB RAM can suffer Elasticsearch out-of-memory kills and silent ingestion failures. source
- Hybrid search
- The product combines vector search, BM25, custom scoring and reranking for retrieval. source
- Intended users
- The site describes the platform as built for enterprise and lists solutions for financial services, legal and compliance, manufacturing, and education. source
- Knowledge compilation
- The platform organizes documents into structured knowledge artifacts through knowledge compilation. source
- License
- The public repository is labeled Apache-2.0 licensed. source
Checked against the maker's pages on 15 Oct 2026.