Change base_url to https://revoyeapi.degird.com/v1, use your Revoye key as the API key, and set the model to revoye/auto or revoye/<provider>. Existing code keeps working, apart from three documented differences.

Python

import os
from openai import OpenAI
 
client = OpenAI(
    base_url="https://revoyeapi.degird.com/v1",
    api_key=os.environ["REVOYE_KEY"],
    timeout=300.0,                      # a browser is doing the work — do not leave this at default
)
 
answer = client.chat.completions.create(
    model="revoye/chatgpt",             # or revoye/auto
    messages=[{"role": "user", "content": "Summarise the CAP theorem."}],
)
 
print(answer.choices[0].message.content)

TypeScript

import OpenAI from "openai";
 
const client = new OpenAI({
  baseURL: `${process.env.REVOYE_API_BASE}/v1`,
  apiKey: process.env.REVOYE_KEY,
  timeout: 300_000,
});
 
const answer = await client.chat.completions.create({
  model: "revoye/auto",
  messages: [{ role: "user", content: "Summarise the CAP theorem." }],
});

The three differences

Messages are flattened. A system message is prepended and prior assistant turns become context in a single prompt. The provider's own UI holds the real thread; if you need to continue an actual conversation, use the native endpoint with conversation_ref and mode: "continue".

stream: true returns 400. There is nothing to stream. Any UI code that assumes a stream needs a non-streaming path before it can talk to Revoye.

usage is null. No token counts exist. Guard any code that does arithmetic on it — cost estimation, budget enforcement, context-window maths — because None + int is the failure you will hit first.

The timeout, again

Every OpenAI SDK ships with a timeout tuned for a model API answering in seconds. Revoye answers in tens of seconds. Set it explicitly to at least 300 seconds or your client will abandon healthy jobs and report a timeout that did not happen on our side.

Model pickers

GET /v1/models returns your enabled providers in OpenAI's list shape, with revoye/auto first. A picker built against an OpenAI client populates itself with no changes.

What you give up

The compatibility layer covers submitting a prompt and getting an answer. Everything Revoye-specific lives on /v1/completions: asynchronous submission, webhooks, per-attempt timeouts, job deadlines, priority, metadata, pinning a specific agent, and continuing a conversation. For new code, use the native endpoint. The compatibility layer is for saving a rewrite, not for building on.

Using an OpenAI SDK with Revoye