Smith::LLM::OpenAIStream
Reads an OpenAI-shaped chat completion stream.
Shared by OpenRouter, OpenAI and Ollama — smith talks to Ollama through its OpenAI-compatible /v1/chat/completions endpoint, so all three speak this format.
Class methods
read(io : IO, default_model : String, cache_usage : Bool = false, &on_delta : String -> Nil) : Response
cache_usage switches the usage frame to the reading that carries
Anthropic cache counters. Off by default: OpenAI reports a
cached_tokens of its own under the same key, billed at its own rate,
and treating it as an Anthropic cache read would misprice it.