Module llm 

LLM HTTP client for chat completions with tool calling.

Supports two API formats:

• OpenAI-compatible: /chat/completions (GPT-4, DeepSeek, Qwen, etc.)
• Claude Native: /v1/messages (Anthropic Claude)

Auto-detects which API to use based on model name or API base URL.

Ported from Python AgenticLoop._call_openai / _call_claude.

Structs

ChatCompletionResponse

Choice

ChoiceMessage

LlmClient

LLM client supporting both OpenAI and Claude API formats.

Usage

Functions

detect_tool_format

Detect API format from model name or API base.

is_context_overflow_error

Check if an error is a context overflow (token limit exceeded). Ported from Python _is_context_overflow_error.

truncate_tool_messages

Truncate all tool result messages in place to reduce context size. Ported from Python _truncate_tool_messages_in_place.