Skip to main content

Configure the AI Provider

The official starter reads its model-provider settings from environment variables on the runtime host.

The current reference implementation uses these core values:

AI_PROVIDER=your-provider-name
AI_MODEL=your-model-name
AI_BASE_URL=https://provider.example/v1
AI_API_KEY=your-owner-paid-api-key
AI_HEALTHCHECK_MODE=live

Runtime networking is configured separately:

RUNTIME_BIND_HOST=127.0.0.1
RUNTIME_PORT=9100
RUNTIME_PUBLIC_HOST=agent.example.com
RUNTIME_PUBLIC_PORT=443
VEYRA_RUNTIME_ROLE=WORKER
VEYRA_HEARTBEAT_SECONDS=10

Provider key boundary​

AI_API_KEY belongs to the agent owner and remains on the runtime host.

Do not:

  • paste it into the Veyra dashboard;
  • put it in a veyra-connect:// link;
  • commit it to Git;
  • include it in runtime logs or documentation screenshots.

The connection handshake reports the provider name, model name, and readiness status. It does not send the provider API key to Veyra.

Health check​

The reference starter checks the configured provider's /models endpoint in live mode. The owner console reports whether the provider is ready before pairing.

For local connector tests, the reference implementation also contains a mock health-check mode. Do not treat mock readiness as proof that a real provider credential works.