Module llm 

LLM wire types — request / response structures and the LlmProvider trait.

This module defines everything needed to communicate with an LLM backend: the provider trait, completion requests and responses, streaming events, model capabilities, and multimodal support (audio, images, video).

Key types

Type	Role
LlmProvider	Trait implemented by each LLM backend (OpenRouter, Anthropic, Ollama…)
CompletionRequest	Single completion request sent to a provider
CompletionResponse	Non-streaming response wrapper
StreamEvent	Incremental events from the provider’s stream
CompletionStream	Alias for Pin<Box<dyn Stream<Item = Result<StreamEvent>>>>
FinishReason	Why generation stopped (stop, length, tool_calls, content_filter)
ToolDefinition	OpenAI-compatible JSON schema for tool calling
ModelCapabilities	Feature flags (tool_calling, images, reasoning, audio…)
TokenUsage	Token consumption reported by the provider
Modality	Output modality: text, audio, or image
TtsRequest, SttRequest, VideoGenRequest	Multimodal operation requests

Structs

AudioOutputConfig

Configuration for audio output in multimodal completions.

CompletionRequest

A single completion request sent to an LLM provider.

CompletionResponse

Non-streaming completion response wrapper.

FrameImage

An image used as a frame reference in video generation.

ImageEditRequest

Parameters for editing one or more images.

ImageGenConfig

Configuration for image generation requests.

ImageGenRequest

Parameters for generating an image via a generative model.

ImageModelCapabilities

Image capability metadata, when provided in a model catalog.

ModelCapabilities

Feature flags describing what a model supports.

ModelInfo

Metadata about an available LLM model, including capabilities, pricing, and context window.

ModelPricing

Pricing information for a model, including input/output token costs and caching discounts.

NativeCompaction

A provider-issued context window to replay without editing its contents.

ReferencePricing

Reference token rates from a separately identified source.

SttRequest

Parameters for a speech-to-text (transcription) request.

TokenUsage

Token usage counts for a single completion request or turn.

ToolDefinition

OpenAI-compatible JSON schema for a tool exposed to the LLM.

TtsRequest

Parameters for a text-to-speech request.

VideoGenRequest

Parameters for generating a video from text prompts or image frames.

Voice

A voice available for text-to-speech generation.

Enums

AudioFormat

Output audio format for TTS responses.

FinishReason

Why the LLM stopped generating (used in stream and non-stream responses).

ImageOutputFormat

Image format requested from a provider.

Modality

Output modality requested in a completion.

ModelCategory

High-level category for filtering models by use-case.

ReferencePricingSource

Source of a model’s reference rates.

StreamEvent

Events emitted incrementally by the provider’s stream.

Traits

LlmProvider

Trait implemented by each LLM backend (OpenRouter, Anthropic, Ollama, etc.).

Functions

mime_from_extension

Infer MIME type from a file path’s extension.

Type Aliases

CompletionStream

Type alias for a stream of completion events from an LLM provider.

ProviderRegistry

Shared registry mapping each crate::model::Provider to its concrete instance.