Omni.Dialects.OllamaChat (Omni v1.2.1)

Copy Markdown View Source

Dialect implementation for the Ollama native chat API wire format.

See Omni.Dialect for the behaviour specification and delta types.

Notable differences

  • Uses NDJSON streaming instead of SSE — each line is a complete JSON object
  • Tool call arguments arrive complete (as a map, not streamed JSON fragments), similar to Google Gemini
  • Thinking content arrives in a message.thinking field alongside message.content
  • Options like max_tokens and temperature are nested under "options" in the request body