Module llm_completion 

Port definition for streaming LLM chat completions.

This module defines the infrastructure contract that the agent loop uses to drive an LLM. The port is intentionally narrow: it speaks domain types (AgentMessage, ToolDefinition, LlmStreamEvent) and hides all vendor wire-format details (OpenAI JSON schemas, SSE framing, HTTP headers, etc.) behind the trait boundary.

Adapter responsibility

A concrete implementation (e.g. in gglib-axum or gglib-proxy) is responsible for:

1. Translating &[AgentMessage] into the vendor’s messages array, serialising ToolCall::arguments (serde_json::Value) into the JSON string form that OpenAI-compatible APIs require.
2. Translating &[ToolDefinition] into the vendor’s tools array.
3. Parsing the streaming SSE response into a sequence of LlmStreamEvent values, accumulating incremental tool-call deltas where necessary.

The agent loop never sees HTTP, never sees reqwest, and never contains a single OpenAI-specific field name.

Traits

LlmCompletionPort

Port that the agent loop uses to drive a streaming LLM.