Deploy to Railway

Deploy PraisonAI to Railway in one command

Deploy your agents to Railway with a starter railway.json and a single deploy command.

Starter templates ship in the git checkout, not in the PyPI wheel (MANIFEST.in excludes infra/). Run create from a monorepo checkout, or set PRAISONAI_STARTERS_ROOT / PRAISONAI_INFRA_ROOT.

Quick Start

1. Scaffold the project

   The starter drops railway.json and agents.yaml into the current directory.

   praisonai deploy create --template railway
   
2. Authenticate with Railway

   Log in with the Railway CLI.

   railway login
   
3. Deploy

   Push the service to Railway.

   praisonai deploy railway
   

---

How It Works

The command shells out to railway up --detach.

---

CloudConfig support (not yet wired)

praisonai deploy railway runs railway up --detach alone — it does not apply any CloudConfig fields yet. Set them in the Railway dashboard until CLI wiring lands (PraisonAI PR #4876).

CloudConfig field	Applied?	Workaround
env_vars	Not applied (names logged only)	Add as service variables in the Railway dashboard
image	Not applied	Set the service source image in Railway settings
cpu	Not applied	Set resources in the Railway dashboard
memory	Not applied	Set resources in the Railway dashboard
region	Not applied	Choose the region in the Railway dashboard
min_instances / max_instances	Not applied	Configure scaling in the Railway dashboard

Every set field is surfaced via logger.warning(...) and echoed in DeployResult.metadata["unapplied"]. Secret names appear; secret values never reach a log line.

Programmatic consumers reading the JSON result of deploy should inspect metadata["unapplied"] to see which fields the provider silently dropped.

---

Command Options

Argument	Default	Description
FILE (positional)	agents.yaml	Agent file to deploy.

---

Starter railway.json

The railway starter deploys the prebuilt image (no Dockerfile needed).

Setting	Value
healthcheckPath	/health
healthcheckTimeout	300
restartPolicyType	ON_FAILURE

Set the service’s source image to ghcr.io/mervinpraison/praisonai in Railway settings (pin a version tag for production). The image already exposes /health on port 8005.

---

Doctor

Check Railway readiness before deploying.

praisonai deploy doctor --provider railway

deploy doctor --provider railway now checks the Railway CLI — previously this reported “Unknown provider”.

---

Best Practices

Pin the image tag for production

Use a released ghcr.io/mervinpraison/praisonai tag instead of latest in the Railway service settings.

Store secrets as Railway variables

Add OPENAI_API_KEY and other secrets as service variables in the Railway dashboard, not in Git.

Destroy from the dashboard

praisonai deploy destroy is not automated for Railway — remove the service in the Railway dashboard or run railway down manually.

---

Related

Deploy Templates

Scaffold the railway starter.

Deploy to Render

Another one-command PaaS target.