Responses API

/ API docs

POST https://api.opencheese.dev/v1/responses is the OpenAI Responses API. Codex uses it, and so do the OpenAI SDKs' responses methods.

A first request

bash

curl https://api.opencheese.dev/v1/responses \
  -H "Authorization: Bearer $OPENCHEESE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "MODEL_ID", "instructions": "Answer in one sentence.", "input": "What is a gateway?"}'

model is required; input is a string or a list of input items, and instructions becomes the system prompt. The answer's text is in output[].content[] items of type output_text:

json

{
  "id": "resp_abc123",
  "object": "response",
  "status": "completed",
  "model": "MODEL_ID",
  "output": [
    {
      "type": "message",
      "role": "assistant",
      "content": [{ "type": "output_text", "text": "A gateway passes your requests on to the models behind it." }]
    }
  ],
  "usage": { "input_tokens": 24, "output_tokens": 13, "total_tokens": 37 }
}

Parameters

Field	Notes
`model`	Required. An `id` from `/v1/models`.
`input`	A string, or `message`, `function_call` and `function_call_output` items.
`instructions`	The system prompt.
`stream`	`true` for server-sent events.
`tools`, `tool_choice`	Function tools, where the model supports tools.
`reasoning.effort`	A level from the model's `supported_reasoning_levels`.
`temperature`, `top_p`, `max_output_tokens`	Optional. If you don't set them, the model's defaults are used.

Conversations are sent in full

Every request carries the whole conversation in input: replay earlier turns as message items (the assistant's with role: "assistant"). previous_response_id, store and text.format are not supported here; for structured output, use Chat Completions with response_format.

json

{
  "model": "MODEL_ID",
  "input": [
    { "type": "message", "role": "user", "content": "Name three cheeses." },
    { "type": "message", "role": "assistant", "content": "Brie, Gouda and Comté." },
    { "type": "message", "role": "user", "content": "Which one melts best?" }
  ]
}

Streaming

bash

curl -N https://api.opencheese.dev/v1/responses \
  -H "Authorization: Bearer $OPENCHEESE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "MODEL_ID", "stream": true, "input": "Count to three."}'

The stream sends the Responses event types: response.created and response.in_progress, then response.output_item.added, response.output_text.delta with the text, the matching .done events, and finally response.completed with the full response and its usage. A request that fails after the stream has started ends with response.failed, whose response.error holds a code and a message.

text

event: response.output_text.delta
data: {"type":"response.output_text.delta","output_index":0,"content_index":0,"delta":"One, two"}

event: response.completed
data: {"type":"response.completed","response":{"status":"completed","usage":{"input_tokens":12,"output_tokens":6,"total_tokens":18}}}

Tools

json

{
  "model": "MODEL_ID",
  "input": "What is the weather in Paris?",
  "tools": [
    {
      "type": "function",
      "name": "get_weather",
      "description": "Current weather for a city",
      "parameters": { "type": "object", "properties": { "city": { "type": "string" } }, "required": ["city"] }
    }
  ]
}

A tool call arrives as a function_call output item with call_id, name and arguments. Send the result back in the next request as a function_call_output item with the same call_id, after the function_call item itself.

Reasoning effort

bash

curl https://api.opencheese.dev/v1/responses \
  -H "Authorization: Bearer $OPENCHEESE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "MODEL_ID", "reasoning": {"effort": "high"}, "input": "Plan a three-day trip to Lyon."}'

If you don't set an effort, the model's default is used. A level the model doesn't offer is refused with unsupported_reasoning_effort, listing the levels it does offer. Reasoning summaries, where the model returns them, arrive as reasoning output items.

Usage

usage reports input_tokens (with input_tokens_details.cached_tokens), output_tokens (with output_tokens_details.reasoning_tokens) and total_tokens.

Python

python

from openai import OpenAI

client = OpenAI(base_url="https://api.opencheese.dev/v1", api_key="YOUR_OPENCHEESE_API_KEY")
response = client.responses.create(model="MODEL_ID", input="Write a haiku about cheese.")
print(response.output_text)

Errors on this endpoint are always an HTTP status with a JSON body, never an assistant reply as on Chat Completions; once a stream has started, they arrive as response.failed. See Errors and rate limits. Add the opencheese-session-uuid header to keep a conversation together.

The cheese has holes. Your infrastructure should not.