LLM.Adapter.OpenAI (LLM v0.1.2)

Copy Markdown View Source

OpenAI Chat Completions adapter.

Handles the wire format for the /v1/chat/completions endpoint. This adapter is also used by OpenRouter, Ollama, Groq, DeepSeek, xAI, and other OpenAI-compatible providers.

Capabilities

  • Streaming via SSE
  • Tool/function calling
  • Extended thinking via reasoning_effort (o-series models)
  • Model listing via /v1/models

Request format

Encodes messages into OpenAI's %{"role" => ..., "content" => ...} format. System prompts become a system message. Tool definitions use the %{"type" => "function", "function" => ...} wrapper.

SSE events

  • data: {"choices": [{"delta": ...}]} — text, thinking, or tool call chunks
  • data: [DONE] — end of stream