About RAGFlow 0.27.2 — Open-Source RAG Engine With Cited Answers
RAGFlow is a retrieval-augmented generation (RAG) engine from InfiniFlow. It reads your documents with a layout-aware parser, splits them into chunks you can inspect, and answers questions with citations back to the passages it used. Agents, knowledge graphs and connectors to outside data sources sit on top. It is released under the Apache 2.0 licence.
This page links to the v0.27.2 release on GitHub: the source archive of the tag, which the Docker setup runs from, and the release’s admin command-line client. Bineret did not build RAGFlow and is not affiliated with InfiniFlow.
What it does
- Deep document understanding. Its DeepDoc parser pulls knowledge out of complicated layouts: Word, slides, Excel, plain text, images, scanned copies, structured data and web pages. MinerU and Docling are available as parsers too.
- Chunking you can see. Template-based chunking, with the chunks shown so a person can check and correct them before anything is answered from them.
- Grounded citations. Answers point to the references they came from.
- Retrieval. Several recall paths combined with fused re-ranking.
- Agents. Agentic workflows with MCP, a Python and JavaScript code executor, and memory.
- Connectors. Data sync from Confluence, S3, Notion, Discord, Google Drive and others, and chat channels including Feishu, Discord, Telegram and Line.
- APIs for building it into other systems.
It needs a real server
The documented minimum is 4 CPU cores, 16 GB of RAM and 50 GB of disk, with Docker 24.0.0 and Docker Compose 2.26.1 or newer. The default stack runs RAGFlow alongside MySQL, Elasticsearch, Redis and MinIO. On Linux, vm.max_map_count must be at least 262144 for Elasticsearch. The prebuilt Docker images are x86 only, so on ARM64 you build your own. The code executor sandbox needs gVisor.
Since v0.22.0 the image is the slim edition, about 2 GB, with no embedding models inside, so it relies on external model and embedding services.
Change the default passwords
The shipped docker/.env uses the same password, infini_rag_flow, for Elasticsearch, MySQL, MinIO and Redis. The admin service’s superuser password defaults to admin. Change all of them before the server is reachable by anyone but you.
What is in this download
- The source archive, 22.5 MB and 5,977 files. Its
dockerfolder holds the compose file and.envthat start the stack. The README asks you to use the tag matching your image so thatentrypoint.shagrees with it, and this archive is exactly that tag. ragflow-cli, the command-line client for RAGFlow’s Admin Service. It connects on port 9381 and manages the system: list services and their status, and create, disable or remove users and change their passwords. Normal use of RAGFlow does not need it.
What 0.27.2 changed
Released on 10 September 2026. The Agentic RAG retrieval framework was refactored for faster reasoning, and knowledge compilation gained runtime settings and a rate limit. New in this release: a Sitemap data source, custom CA certificates for WebDAV, MonkeyOCRv2 and self-hosted PaddleOCR-VL parsing, and EPUB preview. Fixes include images uploaded in chat being ignored by vision models, and a Starlette upgrade for CVE-2026-54283.
When to use something else
RAGFlow earns its weight when documents are messy, such as scans, tables and complex layouts, and answers have to be traceable. If you are one person with a folder of ordinary PDFs, a 16 GB server is a lot; a desktop app such as AnythingLLM is far lighter.
Key capabilities
Built for messy documents
The DeepDoc parser handles complicated layouts, scanned copies, tables, slides and spreadsheets, with MinerU and Docling as alternative parsers. 0.27.2 adds MonkeyOCRv2 and self-hosted PaddleOCR-VL.
Chunks a person can check
Documents are split by chunking templates, and the chunks are visualised so you can review and correct them. Answers then cite the passages they drew on, so a claim can be traced back.
A real server, not a laptop app
4 cores, 16 GB of RAM and 50 GB of disk at minimum, x86 only for the prebuilt images, with MySQL, Elasticsearch, Redis and MinIO running alongside. The image carries no embedding models.
Change four passwords first
docker/.env ships with infini_rag_flow as the password for Elasticsearch, MySQL, MinIO and Redis, and the admin superuser defaults to admin. Replace every one before anyone else can reach the server.
Agents and connectors
Agentic workflows with MCP, a code executor and memory. Data syncs from Confluence, S3, Notion, Discord and Google Drive, and conversations can run through Feishu, Discord, Telegram or Line.
Questions & Answers
What hardware do I need?
The documented minimum is 4 CPU cores, 16 GB of RAM and 50 GB of disk, on an x86 machine running Docker 24.0.0 and Compose 2.26.1 or newer.
Does it run on an ARM machine, such as an Apple Silicon Mac?
Not with the prebuilt images, which are x86 only. On ARM64 you have to build your own Docker image, following RAGFlow's build guide.
Are embedding models included?
No, not since v0.22.0. The image is the slim edition, about 2 GB, and uses external embedding and model services that you configure.
Elasticsearch or Infinity?
Elasticsearch is the default store for full text and vectors. You can switch to Infinity by setting DOC_ENGINE=infinity in docker/.env, but the documented switch runs docker compose down -v, which deletes existing data. Decide before you load documents.
How do I change the web port from 80?
In docker/docker-compose.yml, change the 80:80 mapping to YOUR_PORT:80, then run docker compose -f docker-compose.yml up -d again.
What are the default passwords?
infini_rag_flow for Elasticsearch, MySQL, MinIO and Redis in docker/.env, and admin for the admin service's superuser. Change them before exposing the server.
Do I need ragflow-cli?
No. It is an administration client for monitoring services and managing users. Everyday use happens in the web interface and through the API.
Is Bineret connected to RAGFlow?
No. This page describes RAGFlow and links to files on InfiniFlow's GitHub release.
Tutorials
1. Check the host
At least 4 CPU cores, 16 GB of RAM and 50 GB of disk, an x86 machine, and Docker 24.0.0 with Docker Compose 2.26.1 or newer. Elasticsearch needs vm.max_map_count of at least 262144:
sysctl vm.max_map_count
sudo sysctl -w vm.max_map_count=262144That setting resets on reboot. To keep it, add vm.max_map_count=262144 to /etc/sysctl.conf.
2. Unpack and change the passwords
unzip ragflow-0.27.2.zip
cd ragflow-0.27.2/dockerOpen .env and replace infini_rag_flow in ELASTIC_PASSWORD, MYSQL_PASSWORD, MINIO_PASSWORD and REDIS_PASSWORD with strong passwords of your own. The image it pulls is set in RAGFLOW_IMAGE, already infiniflow/ragflow:v0.27.2.
3. Start the stack
docker compose -f docker-compose.yml up -dTo use a GPU for DeepDoc tasks, add DEVICE=gpu as the first line of .env before starting.
4. Wait until it is really up
docker logs -f docker-ragflow-cpu-1When the log shows Running on all addresses (0.0.0.0), open http://YOUR_SERVER_IP in a browser; the default web port is 80. Logging in earlier can show a "network abnormal" error while services are still starting.
5. Connect a model
Choose a model provider and enter its API key. Defaults for new users can be set in docker/service_conf.yaml.template under user_default_llm; RAGFlow's llm_api_key_setup guide covers the options.
6. Optional: the admin CLI
Enable the Admin Service by adding --enable-adminserver to the RAGFlow service's command in docker-compose.yml, restart, then connect:
pip install ragflow-cli==0.27.2
ragflow-cli -h 127.0.0.1 -p 9381The superuser password defaults to admin; change it straight away. Commands end with a semicolon, for example LIST SERVICES; or LIST USERS;. The release also carries standalone ragflow-cli binaries if you would rather not use pip.
Support
Where support comes from
RAGFlow is developed by InfiniFlow and contributors. Start with the documentation, report bugs as GitHub issues, and follow the security policy for vulnerabilities. The project also runs a Discord community, linked from its README.
What Bineret covers
This page. If a download link is broken or something written here is wrong, tell us and we will fix it. We do not host, install or tune RAGFlow deployments.
No warranty
Apache 2.0 provides the software as-is. Before an upgrade, back up the Docker volumes that hold MySQL, Elasticsearch and MinIO data, since that is where your parsed knowledge lives.
