OpenAI Responses API

Use /openai/v1/responses only when a client or agent tool specifically expects OpenAI Responses API wiring.

The Responses API route exists for clients that expect OpenAI's newer responses wire format, especially agent coding tools and SDKs that no longer use Chat Completions directly.

For normal application chat, start with OpenAI Compatible API. Use this page when a tool explicitly asks for a Responses API base URL or wire_api = "responses".

Status

Beta. The non-streaming route has been smoke-tested locally. Streaming compatibility depends on the client and model path, so verify it in your target account before production use.

Base URL

https://gateway.api.anyint.ai/openai/v1

Routes

Route	Purpose
`POST /responses`	Create a response
`GET /responses/{response_id}`	Fetch a response by ID when supported by the upstream path

Authentication

Authorization: Bearer <ANYINT_API_KEY>
Content-Type: application/json

Minimal request

Request fields

Field	Type	Required	Meaning	Example
`model`	string	Yes	Model ID available to your account. Call Models API first.	`claude-sonnet-4-6`
`input`	string or array	Yes	User input for the response. Use a string for simple prompts or structured input items for compatible clients.	`Reply with exactly: OK`
`stream`	boolean	No	Requests a streaming response when the selected model path and client support it. Verify before production.	`false`
`instructions`	string	No	System-style instruction for the response.	`Be concise.`
`tools`	array	No	Tool definitions for Responses-compatible clients when supported.	`[{"type":"function",...}]`
`tool_choice`	string or object	No	Controls tool use when tools are provided.	`auto`
`temperature`	number	No	Sampling randomness.	`0.7`
`top_p`	number	No	Nucleus sampling control.	`1`
`max_output_tokens`	integer	No	Maximum number of output tokens for Responses-compatible clients.	`512`
`metadata`	object	No	Customer-defined metadata for tracing. Do not put API keys or private credentials here.	`{"workflow":"codex"}`

curl https://gateway.api.anyint.ai/openai/v1/responses \
  -H "Authorization: Bearer $ANYINT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "input": "Reply with exactly: OK"
  }'

Response fields

Field	Meaning
`id`	Response identifier
`object`	Response object type when returned
`status`	Response status when returned by the compatible path
`model`	Model that handled the request
`output`	Structured output items when returned
`output_text`	Flattened text output when returned by the compatible path
`usage`	Token usage when available
`error`	Error object when the compatible path returns a failed response

Fetch a response

Use this only when the response ID is returned and lookup is supported by the selected path.

curl https://gateway.api.anyint.ai/openai/v1/responses/$RESPONSE_ID \
  -H "Authorization: Bearer $ANYINT_API_KEY"

Codex CLI example

[profiles.anyint]
model_provider = "anyint"
model = "<MODEL_ID>"

[model_providers.anyint]
name = "AnyInt"
base_url = "https://gateway.api.anyint.ai/openai/v1"
env_key = "ANYINT_API_KEY"
wire_api = "responses"

Then run:

export ANYINT_API_KEY="your-anyint-api-key"
codex exec --profile anyint "Reply exactly OK."

Common mistakes

Using this route for a normal OpenAI SDK chat integration when /chat/completions would be simpler
Assuming streaming is verified for every agent tool and model
Forgetting that the base URL ends at /openai/v1, not /openai/v1/responses
Hardcoding a model ID before checking Models API

On this page