Developer Docs › Responses API

Responses API

OpenAI's next-generation Responses endpoint, used by default by Codex CLI and similar tools.

Endpoint

POST https://api.onefast.ai/v1/responses
Authorization: Bearer sk-xxx

The Responses API is OpenAI's recommended new endpoint, with reasoning summaries, built-in tools, multi-turn state and more. It is the endpoint Codex CLI uses (see Codex CLI). It works with GPT models.

Basic request

cURL

curl https://api.onefast.ai/v1/responses \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.5",
    "input": "Write a haiku about the sea.",
    "max_output_tokens": 200
  }'

Python

from openai import OpenAI

client = OpenAI(api_key="sk-xxx", base_url="https://api.onefast.ai/v1")
resp = client.responses.create(model="gpt-5.5", input="Write a haiku about the sea.")
print(resp.output_text)

Node.js

import OpenAI from "openai";

const client = new OpenAI({ apiKey: "sk-xxx", baseURL: "https://api.onefast.ai/v1" });
const resp = await client.responses.create({ model: "gpt-5.5", input: "Write a haiku about the sea." });
console.log(resp.output_text);

Common parameters

Parameter Description
model Required. A GPT model
input Required. A string, or an array of messages [{"role": "user", "content": "..."}]
instructions System instructions
max_output_tokens Maximum output tokens (including reasoning tokens)
reasoning Reasoning settings, e.g. {"effort": "high", "summary": "auto"}
tools Function tools and more
stream Stream the response; event types include response.output_text.delta and response.completed
previous_response_id Optional. Continue from a previous turn; see "Multi-turn conversations" below
store Optional. With false this turn is not saved and cannot be referenced later; an explicit value is forwarded to the model as is

Multi-turn conversations

We recommend sending the full history in input on every turn (stateless style) — it's the most reliable on every route:

Python

history = [{"role": "user", "content": "My name is Alex"}]
r1 = client.responses.create(model="gpt-5.5", input=history)
history += [{"role": "assistant", "content": r1.output_text}, {"role": "user", "content": "What's my name?"}]
r2 = client.responses.create(model="gpt-5.5", input=history)
print(r2.output_text)

Continuing with previous_response_idAlpha

Most users do not need this. Sending the full history every turn (the style above) is the most reliable on every route and is recommended.

Some GPT routes are backed by Azure OpenAI resources spread across several regions, so two turns of the same conversation can land on resources in different regions. Conversation records and encrypted reasoning stored by the upstream cannot be read across resources, so relying on upstream continuation directly can fail with errors such as "previous response not found" or encrypted content that cannot be decrypted. On these routes the gateway therefore keeps the conversation history itself: when it receives previous_response_id it expands the history before sending; when encrypted reasoning sent by your client cannot be decrypted on the current resource, the gateway removes that reasoning and retries once automatically. This handling is still in Alpha, and information that existed only in the previous turn's reasoning may not carry over across resources.

You can also send only the new input and point previous_response_id at the previous response ID:

Python

r1 = client.responses.create(model="gpt-5.5", input="My name is Alex")
r2 = client.responses.create(model="gpt-5.5", previous_response_id=r1.id, input="What's my name?")
  • Continuation relies on temporarily stored conversation records, which expire after the conversation has been inactive for a while.
  • The previous turn's instructions and tools are not inherited; pass them again on each turn if you need them.
  • Turns sent with "store": false are not saved and cannot be referenced by previous_response_id later.
  • If the previous turn cannot be found, you get a 400 error such as Previous response with id ... not found; switch to the full-history style above.

For conversations that must persist long-term or across days, keep the history in your own system and use the full-history style.

Unsupported endpoints

The following endpoints always return 404 with error.code set to unsupported_endpoint:

Endpoint Purpose
GET /v1/responses/{id} Retrieve a stored response
DELETE /v1/responses/{id} Delete a response
GET /v1/responses/{id}/input_items List a response's input items
POST /v1/responses/{id}/cancel Cancel a background response 
{"error": {"message": "Retrieving, deleting, listing input items of, or cancelling stored responses is not supported on this endpoint. To continue a conversation, send POST /v1/responses with previous_response_id.", "type": "invalid_request_error", "param": "response_id", "code": "unsupported_endpoint"}}

File uploads (/v1/files) and background mode (background) are not supported either.