Module llm 

LLM provider abstraction.

A provider takes a transcript plus tool specs and returns either free-form content, structured tool calls, or both. The ChatProvider trait is the only thing the agent depends on; everything beyond it (HTTP, retries, mocking) lives in adapters.

Sub-modules

• chat — data types: Completion, ToolSpec, ToolCall, TokenUsage, …
• pricing — ModelPricing, RetryPolicy, and catalog lookup helpers
• openai / anthropic / mock — concrete provider implementations

Re-exports

pub use chat::estimate_tokens;

pub use chat::Completion;

pub use chat::ContextBreakdown;

pub use chat::StreamChunk;

pub use chat::StreamSender;

pub use chat::StructuredRequest;

pub use chat::TokenUsage;

pub use chat::ToolCall;

pub use chat::ToolSpec;

pub use pricing::context_window_tokens_for_model;

pub use pricing::context_window_tokens_for_model_effective;

pub use pricing::default_compact_threshold_chars;

pub use pricing::default_compact_threshold_tokens;

pub use pricing::pricing_for;

pub use pricing::ModelPricing;

pub use pricing::RetryPolicy;

pub use anthropic::AnthropicProvider;

pub use openai::OpenAiProvider;

Modules

anthropic

Anthropic Messages API adapter.

chat

Chat completion data types.

mock

Deterministic LLM for tests and offline development.

openai

OpenAI-compatible chat-completions adapter.

pricing

LLM pricing and retry policy.

search

Tool search engine: keyword-based discovery of deferred tools.

Constants

DEFAULT_MAX_TOKENS

Default per-response output token cap used when neither the RECURSIVE_MAX_TOKENS env var, the active provider preset’s ModelSpec.max_tokens, nor the config-file agent.max_tokens is set.

Traits

ChatProvider

Core trait for LLM chat-completion providers.