render.yaml

# Render Blueprint — deploys the RiskOS AI backend.
# Dashboard: New + -> Blueprint -> select this repo, or create a Web Service
# manually with the same settings.
services:
  - type: web
    name: riskos-ai-api
    runtime: python
    rootDir: backend
    plan: free
    # Train the ML model at build time (artifact is gitignored, so it must be
    # produced here); requirements-dev includes scikit-learn + test deps.
    buildCommand: pip install -r requirements-dev.txt && python -m scripts.train_model
    # Seed is idempotent — it exits immediately if demo users already exist.
    startCommand: python -m scripts.seed && uvicorn app.main:app --host 0.0.0.0 --port $PORT
    healthCheckPath: /health
    envVars:
      - key: PYTHON_VERSION
        value: 3.11.9
      - key: SMS_ENABLED
        value: "false"          # demo-safe: never call Twilio in deployment
      - key: MOCK_AI
        value: "false"          # real Claude when ANTHROPIC_API_KEY is set;
                                # auto-falls back to mock when it isn't (no crash)
      - key: ANTHROPIC_API_KEY
        sync: false             # set in the Render dashboard — never in the repo.
                                # Leave unset to run the deterministic mock provider.
      - key: ANTHROPIC_MODEL
        value: claude-haiku-4-5 # cheapest model ($1/$5 per MTok)
      - key: DATABASE_URL
        value: sqlite:///./riskos.db
      - key: JWT_SECRET
        generateValue: true
      - key: CORS_ORIGINS
        sync: false             # set manually to your Vercel URL after frontend deploy,
                                # e.g. https://riskos-ai.vercel.app,http://localhost:3000