Providers on a worker
Add-on model providers are configured per worker. Where keys are stored, how they are shown, and how the order and limit behaviour work.
Provider configuration lives on the worker, not in the control plane: keys never leave the machine.
Worker UI → AI models:
- Own logins are listed first: each installed agent as
native:<agent>, used by default. - Add-on providers follow, in the order they are tried. Add them with a preset (OpenRouter, NVIDIA NIM, Groq, DeepSeek, OpenAI, Google Gemini, Anthropic, Ollama, LM Studio, other OpenAI-compatible) or an advanced kind (Azure OpenAI, Bedrock, Vertex).
- When a harness reaches its limit: Ask me or Switch automatically.
Keys are stored in the credential store and only ever shown masked, such as sk-ant-••••••••••••1234.
AO_HARNESS_OWN_LOGIN=0 in the worker's environment turns own-login targets off, for workers that must use
configured providers only.
Details per provider: Providers.