
Deploy Unstract (LLM Document Extraction)
LLM document extraction: Prompt Studio, ETL and APIs, Docker runner free
Just deployed
Unstract
Just deployed
Backend
Just deployed
Worker Pipeline
Just deployed
Platform Service
Just deployed
Qdrant
Just deployed
Postgres
Just deployed
X2Text Service
Just deployed
Redis
Just deployed
Worker Executor
Just deployed
RabbitMQ
Just deployed
Frontend
Just deployed
Worker Background
Just deployed
Deploy and Host Unstract on Railway
Unstract is an open source platform that turns unstructured documents (PDFs, scans, emails, images) into structured JSON with LLMs. You design extraction prompts in Prompt Studio, then run them as ETL pipelines between your storage and databases or as REST API deployments.
About Hosting Unstract
This template runs the official Unstract v0.204.0 images as 13 services: a Caddy gateway on one public domain (web app and API on the same origin, replacing upstream's Traefik), the web frontend, the Django backend, the platform and X2Text services, three worker services that run every Postgres queue consumer from the upstream compose file, and pgvector Postgres, Redis, RabbitMQ, MinIO and Qdrant on the private network. The built in login uses a generated password instead of the fixed unstract/unstract default, there is no self signup, and every internal secret (Django, Fernet encryption key, worker API key, database, Redis, RabbitMQ, MinIO, Qdrant) is generated per deploy.
Requires the Hobby or Pro plan: the stack is 13 services, and Trial projects are limited to 5.
Not included: upstream's Docker based tool runner, which starts containers through the host Docker socket and cannot run on Railway. Prompt Studio projects, exported tools, ETL pipelines and API deployments do not need it (since v0.200 they run inside the workers). The two registry tools that still need it, Classifier and Text Extractor, are hidden.
Common Use Cases
- Invoice, receipt and purchase order extraction into JSON with a schema you define in Prompt Studio
- Document ETL: watch a bucket or drive, extract fields, write rows to Postgres, Snowflake, BigQuery and more
- Document extraction API: publish a Prompt Studio project as a REST endpoint for your apps
- Contract and form review with LLM prompts that cite where each value came from
- Self hosted alternative to hosted document AI, keeping files in your own MinIO bucket
Dependencies for Unstract Hosting
- PostgreSQL with pgvector: included (
pgvector/pgvector:0.8.2-pg17, volume) - Redis, RabbitMQ, MinIO: included (MinIO holds every uploaded and processed file, volume)
- Qdrant: included as the vector database for Prompt Studio indexing (volume, API key on)
- LLM, embedding and text extraction providers: not needed to boot; add them as adapters in Settings after you sign in (for example an OpenAI key and an LLMWhisperer key)
Deployment Dependencies
Implementation Details
# Backend service
DEFAULT_AUTH_USERNAME=unstract
DEFAULT_AUTH_PASSWORD=${{secret(24)}}
ENCRYPTION_KEY=${{secret(43)}}=
WEB_APP_ORIGIN_URL=https://${{Unstract.RAILWAY_PUBLIC_DOMAIN}}
INTERNAL_SERVICE_API_KEY=${{secret(48)}}
First steps after deploy:
- Wait until Backend is healthy (first boot runs the database migrations, a few minutes), then open the Unstract service domain.
- Sign in as
unstractwith theDEFAULT_AUTH_PASSWORDvalue from the Backend service variables. - In Settings, add an LLM, an embedding model, a vector DB and a text extractor. For the vector DB pick Qdrant with URL
http://qdrant.railway.internal:6333and theQDRANT__SERVICE__API_KEYvalue from the Qdrant service. - Create a Prompt Studio project, upload a sample document, write prompts, then export it as a tool and use it in a workflow, ETL pipeline or API deployment.
Never change ENCRYPTION_KEY after adding adapters: stored credentials become unreadable. If you add a custom domain, update WEB_APP_ORIGIN_URL and DJANGO_APP_BACKEND_URL on Backend.
Why Deploy Unstract on Railway?
Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.
By deploying Unstract on Railway, you get the whole document extraction platform wired together in one project: gateway, backend, workers, Postgres, Redis, RabbitMQ, MinIO and Qdrant with generated secrets, private networking, health checks and managed SSL, without running Docker Compose or a reverse proxy yourself.
Template Content
Unstract
caddy:2.11.8-alpineBackend
unstract/backend:v0.204.0Worker Pipeline
unstract/worker-unified:v0.204.0Platform Service
unstract/platform-service:v0.204.0Qdrant
qdrant/qdrant:v1.16.1Postgres
pgvector/pgvector:0.8.2-pg17X2Text Service
unstract/x2text-service:v0.204.0Redis
redis:7.4.6-alpineWorker Executor
unstract/worker-unified:v0.204.0RabbitMQ
rabbitmq:4.1.0-managementFrontend
unstract/frontend:v0.204.0Worker Background
unstract/worker-unified:v0.204.0
