Module model 

Model identifiers, provider variants, and sampling parameters.

The Model type parses from a "provider:model-name" string via FromStr (e.g., "openrouter:anthropic/claude-sonnet-4") and serialises back to the same format. Aliased models ("fast,gpt-4.1") let users define shorthand names that resolve to real model IDs at API-call time.

Key types

Type	Role
Model	Fully-qualified model ID (provider:name); parses bidirectionally
Provider	LLM backend variant (OpenRouter, Ollama, Anthropic, OpenAI, Connect)
AliasedModel	Model name with optional aliasing (name,real_name)
ReasoningLevel	Extended chain-of-thought effort (Off / Low / Medium / High)
SamplingParams	Extended sampling config (top_p, top_k, min_p, penalties)
AgentLlmConfig	Complete LLM config for an agent turn (model, temp, reasoning, sampling)

Structs

AgentLlmConfig

Complete LLM configuration for an agent turn.

AliasedModel

A model name with optional aliasing.

Model

Fully-qualified model identifier combining a backend Provider with an AliasedModel.

ProviderPreferences

Provider-level routing preferences (OpenRouter extension).

SamplingParams

Extended sampling parameters beyond just temperature.

Enums

ApiKey

An API key identifier, not a credential value.

Provider

LLM backend variant. Used to route completion requests and select the correct crate::llm::LlmProvider implementation from the registry.

ProviderSort

How to sort provider candidates.

Quantization

Quantization level for model weights.

ReasoningLevel

Extended chain-of-thought / reasoning effort level.