Module model 

The chat-model abstraction the agent loop runs against.

Decoupling the loop from a concrete client means: (a) the loop is unit- testable with a scripted mock — no network; (b) any backend (the real LlmClient, a local model, a replay harness) plugs in by implementing one method.

Structs

ModelDescriptor

Registered model and the request options it supports.

ModelRegistry

Exact-id registry of models a Profile may select.

ProviderCapabilities

Context-runtime features a provider can carry. Every switch defaults to enabled; an adapter turns one off when its provider cannot honour it, and a host may override the whole set on the runtime.

Enums

ModelRegistryError

Model registration or resolution failure.

Traits

ChatModel

Anything that can turn a chat-completion request into a response.