Responses - airmux

POST /inf/v1/responses

The route always returns Responses-shaped output and errors.

Request

curl https://llm.example.com/inf/v1/responses \
  -H "Authorization: Bearer $AIRMUX_INFERENCE_KEY" \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "openai/gpt-4o-mini",
    "input": "Explain request routing in one sentence",
    "max_output_tokens": 128,
    "store": false
  }'

Supported fields

Field	Notes
`model`	Required catalog model ID
`input`	String or supported input-item array
`instructions`	Prepended as a system message
`stream`	Boolean
`max_output_tokens`	Maximum generated tokens; maps to the gateway’s internal output-token limit
`temperature`, `top_p`	Sampling fields
`tools`	Function tools only; hosted tools are rejected
`tool_choice`	`auto`, `none`, `required`, or named function
`parallel_tool_calls`	Boolean
`reasoning`	Reasoning configuration
`text.format`	`text`, `json_object`, or `json_schema`
`store`	Only `false` or omitted is supported
`include`	Omitted, empty, or `reasoning.encrypted_content`

Unknown top-level fields are rejected with unsupported_feature on this surface.

Response

The buffered response contains output items for assistant messages, reasoning, and function calls. It also includes usage plus an additive gateway object.

{
  "id": "019...",
  "object": "response",
  "status": "completed",
  "model": "openai/gpt-4o-mini",
  "output": [
    {
      "type": "message",
      "id": "msg_0",
      "role": "assistant",
      "status": "completed",
      "content": [{"type": "output_text", "text": "Requests are routed by catalog model, policy, and eligible credential."}]
    }
  ],
  "gateway": {"finish_reason": "stop", "adjustments": []}
}

Streaming

The stream emits named SSE events such as response.created, response.output_item.added, text or function-argument deltas, response.output_item.done, and response.completed. A failure after streaming begins is emitted as an error event.