Models and routing
Maestro selects a provider by capability first: planning, coding, testing, reviewing, research, conversation or improvement review. A provider may expose a model list through its CLI or endpoint, or require a manually configured model.
Preferred models
Section titled “Preferred models”The provider control plane can store a preferred model for a provider/capability. The selected model is applied to the primary provider; a fallback provider is not given an incompatible model ID by accident. If no model is configured, the provider’s account/default model is used.
Model discovery is best-effort. A provider can be healthy while model discovery is unavailable, and the UI should distinguish “no model list returned” from “provider not authenticated.”
Capability order
Section titled “Capability order”Use Providers → Priority by function to choose the order for planning, implementation, tests, final review, research and conversation. “Use only this provider” pauses the others and disables fallback for the selected provider; use it as a temporary control, not as a guarantee that the provider will accept every request.
Quota and billing
Section titled “Quota and billing”Maestro does not create a new OpenAI billing surface. Provider subscriptions, API billing and quota rules remain external. See Cost and quota.