Migrate an OpenAI-compatible client

The snippets below adapt an existing client and assume its messages conversation already exists. For a standalone program, use First request.

1. Change three settings

import os
from openai import OpenAI

client = OpenAI(
    base_url=os.environ["TOKENHUB_BASE_URL"],  # https://tokenhub-api.belive.sg/api/v1
    api_key=os.environ["TOKENHUB_API_KEY"],
)
r = client.chat.completions.create(model=os.environ["TOKENHUB_MODEL"], messages=messages)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: process.env.TOKENHUB_BASE_URL, // https://tokenhub-api.belive.sg/api/v1
  apiKey: process.env.TOKENHUB_API_KEY,
});
const r = await client.chat.completions.create({ model: process.env.TOKENHUB_MODEL!, messages });

2. Check what changes

• The base URL ends in /api/v1. Do not add another /v1: the SDK appends /chat/completions itself.
• Model ids carry an author, such as deepseek/… or qwen/…. A bare OpenAI model name returns 404.
• Check each option you send against the model's supported_parameters. Provider routing fields such as provider.only work; an unknown key inside provider returns 400 unsupported_parameter instead of being dropped.
• Streams end with data: [DONE] and carry usage on a final chunk with an empty choices array. See Terminal chat.
• Errors keep the OpenAI shape and a stable type; see Errors for which are safe to retry.

3. Leave everything else alone

Keep your other providers, retries and logging as they are. A migration is done when one request and one streamed request succeed with the new settings and your existing tests pass.