Responses

POST /<lane>/v1/responses — the OpenAI Responses API. This is the only wire protocol Codex supports (wire_api = "responses"), and the OpenAI SDK can use it via client.responses.create.

Request

curl "https://api.opengateway.one/oss/v1/responses" \
  -H "Authorization: Bearer $OPENGATEWAY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-turbo-hub-gpt-oss-120b",
    "input": "Say PONG only.",
    "reasoning": { "effort": "low" }
  }'

Response

{
  "id": "resp_...",
  "object": "response",
  "model": "claude-turbo-hub-gpt-oss-120b",
  "output": [
    {
      "type": "message",
      "role": "assistant",
      "content": [{ "type": "output_text", "text": "PONG" }]
    }
  ],
  "usage": { "input_tokens": 5, "output_tokens": 1, "total_tokens": 6 }
}

Streaming & effort

• Set stream: true for SSE response events.
• reasoning.effort accepts none|minimal|low|medium|high|xhigh — Codex maps its model_reasoning_effort onto this. See Effort levels.
• previous_response_id continuity is supported for multi-turn flows.

See Codex setup for the full ~/.codex/config.toml.