RAGFlow Document AI Engine

Deploys RAGFlow v0.27, an open-source retrieval-augmented generation engine that parses PDFs, scans and tables into chunks and answers with citations. Includes a separate parsing worker, Elasticsearch, MySQL, Redis and a Railway Bucket.

What you get

  • RAGFlow web service (UI, API, admin server) and a separate ragflow-worker for document parsing, image v0.27.2
  • Elasticsearch 8.11.3 for chunks, full-text and vector search, on a volume
  • MySQL 9 and Redis 8.2, each on a volume
  • A Railway Bucket for uploaded files through RAGFlow's S3 backend instead of MinIO
  • Generated passwords and secrets; parsing capacity scales with worker replicas

Good for

  • Chatting with contracts, policies, manuals and research papers with cited source passages
  • Building a knowledge base over scanned PDFs, tables and slide decks
  • Exposing a retrieval API or MCP endpoint for AI agents
  • Agent workflows that combine document retrieval with web search and tools

Services in this template

  • ragflow · from baranberkay96/ragflow-railway
  • ragflow-worker · from baranberkay96/ragflow-railway
  • Redis · image redis:8.2
  • MySQL · image mysql:9
  • Elasticsearch · from baranberkay96/ragflow-railway

Questions people ask

How do I log in, and where do model keys go?

Wait for the ragflow healthcheck to pass (the first start can take several minutes), open the public URL and register. Registration is open by default; set REGISTER_ENABLED to 0 on the ragflow service afterwards. Add your chat and embedding model providers in the UI. The built-in admin@ragflow.io user uses the generated ADMIN_DEFAULT_PASSWORD from the ragflow variables.

What drives the cost?

RAGFlow is a memory-heavy stack. The main consumers are Elasticsearch, the parsing worker (each executor needs roughly 1.5 to 3 GB while parsing) and the RAGFlow web service. The template's own planning estimate is roughly 6 to 10 GB of RAM in total; this is an estimate, not a measurement.

Is my data kept across redeploys?

Yes. Elasticsearch, MySQL and Redis use volumes, and original files are stored in the Railway Bucket.

How do I upgrade RAGFlow?

The version is the FROM line in services/ragflow/Dockerfile, shared by both RAGFlow services. Read the release notes first; database migrations run on start. Keep Elasticsearch on the version RAGFlow's own docker settings specify.

Are there licence notes?

RAGFlow is Apache-2.0. Elasticsearch 8.11 is distributed under Elastic License 2.0 or SSPL, and running it unmodified as a backing service is permitted under ELv2. MySQL Community is GPL-2.0 and Redis 8 is RSALv2, SSPLv1 or AGPLv3.

Full template README

Deploy and Host RAGFlow with Railway

RAGFlow is an open-source retrieval-augmented generation engine built around deep document understanding: it parses PDFs, scans, tables and slides into well-structured chunks and answers questions with traceable citations. This community template deploys RAGFlow v0.27 on Railway with Elasticsearch, MySQL, Redis, a Railway Bucket and a separate parsing worker.

About Hosting RAGFlow

RAGFlow combines several components. The web service serves the UI and API through nginx; task executors run the DeepDoc OCR and layout models that turn uploaded files into chunks, call your embedding model and write chunks and vectors to Elasticsearch. MySQL stores users, datasets and settings, Redis carries the task queue, and object storage keeps the original files. Parsing is CPU- and memory-intensive, so this template runs it in its own service that you can scale independently of the UI. Original files go to a Railway Bucket through RAGFlow's S3 backend instead of a MinIO container, and every password and secret is generated at deploy time. Expect a memory-heavy stack: plan for roughly 6–10 GB of RAM.

Common Use Cases

  • Chat with contracts, policies, manuals and research papers, with answers that cite the exact source passages.
  • Build an internal knowledge base over scanned PDFs, tables and slide decks that plain text extractors handle poorly.
  • Expose a retrieval API or MCP endpoint for AI agents and assistants.
  • Create agent workflows that combine document retrieval with web search and tools.

Dependencies for RAGFlow Hosting

  • RAGFlow CPU image infiniflow/ragflow:v0.27.2 (web and worker roles)
  • Elasticsearch 8.11.3 (the version RAGFlow pins)
  • MySQL (Railway MySQL 9) and Redis (Railway Redis 8.2)
  • S3-compatible object storage (Railway Bucket)
  • An LLM and an embedding model provider (OpenAI, Azure, Gemini, Ollama, a LiteLLM gateway, ...), configured in the UI

Deployment Dependencies

Implementation Details

Service Role Public Storage
ragflow nginx + UI, API server, admin server, data-source sync Yes (port 80) –
ragflow-worker Document parsing task executors No –
Elasticsearch Chunks, full-text and vector index No Volume
MySQL Users, datasets, documents, settings No Volume
Redis Task queue, locks, caches No Volume
Bucket Uploaded files and parsed images – Railway Bucket

First login

  1. Wait for the ragflow healthcheck (/api/v1/system/healthz) to pass; the first start initializes the database and can take several minutes.
  2. Open the public URL and register your account. Registration is open by default (REGISTER_ENABLED=1) so you can create the first users; set it to 0 on the ragflow service afterwards.
  3. Under Model providers, add your chat and embedding models. Then create a dataset, upload files and start parsing.
  4. The built-in superuser admin@ragflow.io (admin server and CLI) uses the generated ADMIN_DEFAULT_PASSWORD from the ragflow variables.

Scaling: add replicas to ragflow-worker, or raise RAGFLOW_TASK_EXECUTORS, to parse more documents in parallel (each executor needs roughly 1.5–3 GB of RAM while parsing). Raise ES_JAVA_OPTS on Elasticsearch for large datasets.

Pinning and upgrades: the RAGFlow version is the FROM line in services/ragflow/Dockerfile (shared by both RAGFlow services). Read the release notes before upgrading; database migrations run on start. Elasticsearch is pinned in services/elasticsearch/Dockerfile; keep it on the version RAGFlow's docker/.env specifies.

Why Deploy RAGFlow on Railway?

Railway runs the whole stack as one project with private networking, volumes, a managed S3 bucket and generated credentials, so you can stand up a production-shaped RAGFlow without assembling Docker Compose on a VM. Parsing capacity scales with replicas, and you pay for the memory and CPU the stack actually uses.

Guides for this template

View RAGFlow Document AI Engine on railway.com

Similar templates

TemplateCategoryDeploymentsDeploys
9Router
Self-hosted AI router for Claude, Gemini, Cursor & OpenAI tools.
AI/ML9,734
flowise-railway
Flowise - low-code LLM apps builder
AI/ML8,512
LibreChat
Open-source ChatGPT clone: multi-model, agents, search, multi-user auth
AI/ML6,762
Hermes Agent by Nous Research
Template for Hermes Agent by Nous Research
AI/ML6,413
OpenClaw
[Homebrew support] The AI that actually does things
AI/ML5,167
Hermes Agent
Self Host Hermes Agent. Autonomous AI agent w/ self-improving, tools & more
AI/ML4,092