Responses API

POST https://dev.typno.ai/v1/responses accepts OpenAI's Responses format, which Codex and some newer SDKs use. Typno AI translates each request to a chat completion and back.

curl https://dev.typno.ai/v1/responses \
  -H "Authorization: Bearer $TYPNO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "mistral4-119b", "input": "Say hello in Dutch."}'

What works

• input as text or as a list of messages, function calls and function call outputs
• instructions, max_output_tokens, temperature, top_p
• tools of type function and custom, with tool_choice and parallel_tool_calls
• stream: true, with the usual response.* events
• images as input_image parts on models with vision

What does not

Nothing is stored between requests, so previous_response_id and store are not supported: send the whole conversation in input. Hosted tools (web search, file search, code interpreter) do not exist here and are left out of the request.