Grok Responses

Use the Grok / xAI API family for xAI-compatible Responses requests. The Grok / xAI route restricts dispatch to Grok-family models and upstreams; the HTTP request and SSE response bodies use xAI’s OpenAI-compatible Responses shape.

This protocol describes the routing boundary, not a different customer payload format. An OpenAI SDK can use this API when its base_url is set to https://api.hao.ai/grok/v1.

Base URL and endpoint

Protocol root:


https://api.hao.ai/grok

Responses endpoint:


POST https://api.hao.ai/grok/v1/responses

Authentication

Pass your HaoAI API key as a Bearer token:


Authorization: Bearer <your HAOAI_API_KEY>

Request examples

cURL

Terminal


curl https://api.hao.ai/grok/v1/responses \
  -H "Authorization: Bearer $HAOAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok/grok-4.5",
    "input": "Explain how an API gateway works.",
    "instructions": "You are a helpful technical assistant.",
    "max_output_tokens": 1024
  }'

Python

grok_responses.py


from openai import OpenAI
 
client = OpenAI(
    base_url="https://api.hao.ai/grok/v1",
    api_key="<your HAOAI_API_KEY>",
)
 
response = client.responses.create(
    model="grok/grok-4.5",
    input="Explain how an API gateway works.",
    instructions="You are a helpful technical assistant.",
    max_output_tokens=1024,
)
 
print(response.output_text)

TypeScript

grok-responses.ts


import OpenAI from 'openai'
 
const client = new OpenAI({
  baseURL: 'https://api.hao.ai/grok/v1',
  apiKey: '<your HAOAI_API_KEY>'
})
 
const response = await client.responses.create({
  model: 'grok/grok-4.5',
  input: 'Explain how an API gateway works.',
  instructions: 'You are a helpful technical assistant.',
  max_output_tokens: 1024
})
 
console.log(response.output_text)

Streaming

Set stream: true to receive xAI-compatible SSE events:

grok_responses_stream.py


stream = client.responses.create(
    model="grok/grok-4.5",
    input="Write a short welcome message.",
    stream=True,
)
 
for event in stream:
    if event.type == "response.output_text.delta":
        print(event.delta, end="", flush=True)

Continuations and WebSocket clients

Use previous_response_id only with a response previously created by the same HaoAI principal. The router keeps the continuation bound to its original Grok upstream account; unavailable or foreign response IDs are rejected instead of being sent to another account.

A compatible Responses WebSocket ingress is available at:


GET https://api.hao.ai/grok/v1/responses

It is a HaoAI bridge to xAI HTTP/SSE for each turn. It does not indicate that xAI provides a native WebSocket transport.

Current text models

Model ID	Notes
`grok/grok-4.5`	Default frontier model with a 500,000-token context window.
`grok/grok-composer-2.5-fast`	Text-only coding model with a 200,000-token context window; image input is not supported.

For Chat Completions instead of the Responses shape, see Grok Chat Completions.