render.yaml

# Render Blueprint for the backend (free plan). See docs/deployment.md.
# Values marked sync: false are secrets or deployment-specific: Render asks for them when the
# Blueprint is created, and they are never stored in this file.
services:
  - type: web
    name: ai-sql-analyst-api
    runtime: docker
    plan: free
    # Pick the region closest to your Neon database (oregon, ohio, virginia, frankfurt, singapore).
    region: ohio
    dockerfilePath: ./backend/Dockerfile
    dockerContext: ./backend
    healthCheckPath: /api/health
    autoDeployTrigger: commit
    buildFilter:
      paths:
        - backend/**
    envVars:
      # Neon's DIRECT connection string as the read-only sql_agent role (not the owner, and not
      # the -pooler host: the pooler rejects the per-session options the app sets).
      - key: DATABASE_URL
        sync: false
      - key: LLM_API_KEY
        sync: false
      # JSON list with the Vercel origin, e.g. ["https://<app>.vercel.app"]
      - key: CORS_ORIGINS
        sync: false
      - key: LLM_BASE_URL
        value: https://api.groq.com/openai/v1
      - key: LLM_MODEL
        value: openai/gpt-oss-120b
      # Never true on a public deployment: it lets any visitor make the server connect anywhere.
      - key: ALLOW_UI_CONNECTIONS
        value: "false"
      # Sized for Groq's free tier for gpt-oss-120b (8K tokens/minute, 200K tokens/day).
      - key: RATE_LIMIT_PER_MINUTE
        value: "3"
      - key: RATE_LIMIT_PER_DAY
        value: "40"
      - key: TRUSTED_PROXY_HOPS
        value: "0"
      - key: ANSWER_MODE
        value: llm