Module client 

HTTP streaming client for LLM APIs.

Sends conversation messages to an LLM API and streams back response events via Server-Sent Events (SSE). Features:

• Prompt caching with cache_control markers
• Beta header negotiation (thinking, structured outputs, effort)
• Retry with exponential backoff and fallback model
• Tool choice constraints
• Thinking/reasoning token configuration

Structs

CompletionRequest

A request to the LLM API.

LlmClient

Client for communicating with an LLM API.

Enums

EffortLevel

Agent effort level (influences thoroughness and token usage).

ThinkingMode

Configuration for thinking/reasoning behavior.

ToolChoice

Controls how the model selects tools.