Model selection

Request.Model is a complete descriptor, not a string model name. It lets a client select provider namespace, API dialect, endpoint, model ID, capability gates, limits, and defaults without carrying a secret.

API surface

request := inference.Request{
	Model: model.CustomModel(
		model.ProviderName("acme"),
		model.APIFormatOpenAI,
		"https://api.example.test/v1",
		"chat-1",
		model.WithTools(),
	),
}
if err := request.Model.Validate(); err != nil {
	panic(err)
}

Model.BaseURL may be empty as a wildcard for a later trust boundary. A non-empty URL must be HTTPS, or HTTP to loopback only. The client or integration layer resolves any catalogue policy and applies authentication separately.

Stable identity

Use request.Model.Key() when caching or attributing results. ModelKey contains Provider and Model only, so it remains stable if an endpoint or codec route changes. Call Key().Validate() when an empty provider or name would be unsafe.

Proof

• Source: inference/client.go, inference/model/model.go
• Tests: inference/model/model_test.go, inference/model/modelkey_test.go

Related: Model, Model identity.