Module llm 

LLM client — hybrid provider with trait-based adapter.

LlmProvider trait with multiple implementations:

• OllamaProvider — local Ollama server
• OpenAiProvider — OpenAI compatible APIs

Structs

Message

A message in the conversation.

OllamaProvider

Ollama LLM provider.

OpenAiProvider

OpenAI-compatible provider (works with OpenAI, OpenRouter, etc.)

ProposedToolCall

A tool call the model proposed in its response. Awareness ≠ permission: a proposed call is not executed here — the caller resolves it to a route and runs it through the consent/audit path.

ProviderConfig

Configuration for LLM provider selection.

Response

Complete LLM response.

ResponseChunk

LLM response chunk (for streaming).

ToolDef

A tool the model may call, in the provider-agnostic shape the kernel hands down the tools channel. parameters is a JSON Schema object describing the call arguments (the same input_schema a intent::ToolDescriptor carries). Producers route any untrusted description through intent::sanitization before it reaches a provider.

Usage

Token usage statistics.

Enums

LlmError

Errors from the LLM layer.

Role

Message roles.

Traits

LlmProvider

Trait for LLM providers.

Functions

build_failover_chain

Build a failover chain from all configured providers.

create_provider

Create an LLM provider from configuration.

extract_json_from_response

Extract a JSON object from an LLM response string.

select_provider

Probe every configured provider, pick the first reachable one whose preferred_models intersects the live model list, and return it.