What is an OpenAI-compatible API, and how do I use one?

It means that a service implements request and response conventions used by supported OpenAI-style operations. For example, a chat request may accept a model name and a messages array, then return choices containing a message.

The term is not a promise that every SDK method, model parameter or native feature will work. Check the specific endpoint, authentication format and supported model capabilities. Orux's base URL is https://orux.top/api/v1; its documented operations and model forms are the source of truth for integration.

Create an Orux API key, choose an available chat model from the catalog, and set ORUX_API_KEY and ORUX_MODEL in your server environment. This example uses only Python's standard library. Running it makes a real API request and may consume Credits.

import json
import os
import urllib.request

payload = {
    "model": os.environ["ORUX_MODEL"],
    "messages": [{"role": "user", "content": "Reply with a short greeting."}],
    "stream": False,
}
request = urllib.request.Request(
    "https://orux.top/api/v1/chat/completions",
    data=json.dumps(payload).encode("utf-8"),
    headers={
        "Authorization": "Bearer " + os.environ["ORUX_API_KEY"],
        "Content-Type": "application/json",
    },
    method="POST",
)
with urllib.request.urlopen(request, timeout=120) as response:
    result = json.load(response)
print(result["choices"][0]["message"]["content"])

This is a non-streaming chat example. Image, video and audio models may need different operations and input fields. See the API documentation for those requests.

For supported OpenAI-compatible operations, configure the client's base URL as https://orux.top/api/v1, use your Orux API key, and select an Orux model alias. Then test the exact SDK method and payload you intend to use.

Do not assume that changing the base URL makes every OpenAI-specific feature portable. Validate streaming event handling, tool calls and structured-output behavior with your selected model. If your application depends on an unsupported native endpoint, keep that integration separate. Orux is an independent service, not an OpenAI account or subscription.

First check that the model supports the endpoint and every input field you sent. Common differences include input media types, allowed resolution or duration values, context limits and synchronous versus asynchronous execution.

Inspect the error response and your request record before retrying. Correct an invalid payload instead of repeatedly sending it. When a request may already have been accepted, check its recorded outcome before issuing another generation. For asynchronous results, continue polling the original task identifier.