Deploy and Host OpenMed on Railway

OpenMed is an open-source toolkit for clinical text: it finds diseases, drugs, genes and anatomy in medical notes, and it detects and removes personal information (names, dates of birth, record numbers, phones, addresses) in many languages. Its models are small open models that run on CPU. This template runs OpenMed's REST API from a pinned release, protected by an API key from the first boot, with downloaded models kept on a volume.

About Hosting OpenMed

One service, the official OpenMed image, with a volume for the model cache. Models download from Hugging Face the first time a request uses them (a few seconds), then load from the volume after every redeploy.

OpenMed's own API key auth is turned on: every model route needs the X-API-Key header, while /health, /readyz and the /docs page stay open. The key is generated when you deploy. Request logs never contain the text you send.

Common Use Cases

De-identifying clinical notes before they reach an LLM, a data warehouse or a vendor
Extracting diagnoses, medications and other entities from free-text medical records
A PII-redaction API for support tickets, transcripts or forms, in English and other languages
A self-hosted, open alternative to cloud medical NLP services, with data that stays in your project

Dependencies for OpenMed Hosting

Nothing external. Models come from Hugging Face; set HF_TOKEN only for gated models or heavy use.

Deployment Dependencies

Implementation Details

First use: copy OPENMED_API_KEY from the service's Variables tab, then:

curl -X POST https://‹your-domain›/pii/deidentify \
  -H "X-API-Key: $OPENMED_API_KEY" -H "Content-Type: application/json" \
  -d '{"text":"Patient John Smith, DOB 03/14/1962, phone 555-201-3344.","lang":"en","method":"mask"}'
# "Patient [first_name] [last_name], DOB [date_of_birth], phone [phone_number]."

Medical entities:

curl -X POST https://‹your-domain›/analyze \
  -H "X-API-Key: $OPENMED_API_KEY" -H "Content-Type: application/json" \
  -d '{"text":"Type 2 diabetes and hypertension, started on metformin.","model_name":"disease_detection_superclinical"}'

Open https://‹your-domain›/docs for every endpoint: /pii/extract, /pii/deidentify (mask, remove, replace or hash), /analyze, /brief, /ground, streaming and async jobs.

Models and memory: idle uses about 200 MB. The default English PII model brings it to about 350 MB; a large entity model (434M parameters) adds about 2 GB, so give the service 2.5 GB or more for those. OPENMED_SERVICE_KEEP_ALIVE unloads idle models, and OPENMED_SERVICE_MAX_RESIDENT_MODELS caps how many stay loaded. List models in OPENMED_SERVICE_PRELOAD_MODELS to load them at boot, so the first request isn't slow.

Speed: OMP_NUM_THREADS caps CPU threads per call. Without it torch starts one thread per host core, and a short request takes seconds instead of a quarter of one.

Custom domain: add it to OPENMED_SERVICE_TRUSTED_HOSTS, or OpenMed rejects requests to it.

Notes and limits:

Change OPENMED_API_KEY and redeploy to rotate the key.
OpenMed is a research toolkit, not a certified medical device. Review de-identified output before treating it as anonymous.
Each template update pins a newer OpenMed release.

Why Deploy OpenMed 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 OpenMed on Railway, you are one step closer to supporting a complete full-stack application with minimal burden. Host your servers, databases, AI agents, and more on Railway.