Module llm 

LLM client abstraction layer

Provides a unified interface for interacting with LLM providers (Anthropic Claude, OpenAI, Zhipu AI GLM, and OpenAI-compatible providers).

Re-exports

pub use anthropic::AnthropicClient;

pub use codex_login::CodexLoginClient;

pub use factory::create_client_with_config;

pub use factory::LlmConfig;

pub use http::clear_http_metrics_callback;

pub use http::default_http_client;

pub use http::set_http_metrics_callback;

pub use http::HttpClient;

pub use http::HttpClientError;

pub use http::HttpMetricsCallback;

pub use http::HttpMetricsRecord;

pub use http::HttpResponse;

pub use http::StreamingHttpResponse;

pub use openai::OpenAiClient;

pub use zhipu::ZhipuClient;

Modules

anthropic

Anthropic Claude LLM client

codex_login

factory

LLM client factory

http

HTTP utilities and abstraction for LLM API calls

openai

OpenAI-compatible LLM client

structured

Structured object generation from LLM output.

zhipu

Zhipu AI (GLM) LLM client

Structs

Attachment

Image attachment for multi-modal messages.

ImageSource

Image source for the ContentBlock::Image variant.

LlmResponse

LlmResponseMeta

LLM response

Message

Message in conversation

ModelGenerationAdmission

Shared admission gate derived from a typed provider concurrency contract.

ModelGenerationConcurrency

Bounded active model-generation capacity reported by an LlmClient.

ModelGenerationPermit

Owned capacity for one active model-generation transaction.

ModelGenerationPool

Digest-only provider/model capacity metadata.

ModelGenerationPoolHealthSnapshot

Read-only, secret-free configuration and occupancy evidence for one model generation pool.

NonRetryableLlmError

RetryConfig

Configuration for API retry behavior

RetryExhaustedError

A retry loop that exhausted a typed HTTP response status.

SecretString

A string wrapper that redacts its value in Debug and Display output. Prevents API keys from leaking into logs and error messages.

TokenLogProb

Token-level log probability emitted by an OpenAI-compatible backend.

TokenUsage

Token usage statistics

ToolCall

Tool call from LLM

ToolDefinition

Tool definition for LLM

TopTokenLogProb

Enums

ContentBlock

Message content types

ModelGenerationAdmissionError

ModelGenerationPoolError

Errors returned while deriving a provider/model capacity pool identity.

StreamEvent

Streaming event from LLM

ToolResultContent

Content within a tool result — either text or an image.

ToolResultContentField

The content field of a ToolResult block.

ToolResultTrustV1

Typed trust label for tool-result content at the value boundary (KRN-5).

Constants

MAX_ATTACHMENT_BYTES

Maximum bytes loaded by Attachment::from_file.

MAX_RETRIES

Hard upper bound for provider retries accepted by the Core loop.

Traits

LlmClient

LLM client trait