Point the OpenAI Python SDK at Refuseless

You switch the OpenAI Python SDK to Refuseless by passing base_url="https://api.refuseless.com/v1" when you construct the client, and by sending a Refuseless model id. The rest of chat.completions.create stays the shape you already have. The key is a Refuseless API key from the console, not an OpenAI key. The SDK will happily send an OpenAI key to the wrong host, and you will get a 401.

Install and call

Install the official SDK (pip install openai) if the project does not have it. Export REFUSELESS_API_KEY. Then construct the client with the base URL. The sample uses glm-5.3-refuseless, the id used for GLM 5.3 Abliterated in this codebase. If the models page lists a different id for that checkpoint, copy that id. GET /v1/models returns the ids the key can call.

from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["REFUSELESS_API_KEY"],
    base_url="https://api.refuseless.com/v1",
)

response = client.chat.completions.create(
    model="glm-5.3-refuseless",
    messages=[
        {"role": "user", "content": "Summarize this function in one paragraph."},
    ],
)
print(response.choices[0].message.content)

What you do not change

• Message roles. system, user, and assistant mean what they mean in the OpenAI shape.
• The path. The SDK appends /chat/completions to the base URL. Do not also append /v1 in the base URL and in the path.
• Streaming, if you already pass stream=True. Leave it. This page does not add a second client.

Errors you will actually see

• HTTP 401, "Missing bearer API key." The Authorization header never arrived. Check that the environment variable is set in the process that imports the SDK.
• HTTP 401, "Invalid API key." The header arrived and the key is not one of ours. An OpenAI key will do this.
• HTTP 404, code model_not_found, message "Unknown model." The id is not in the catalog. Copy one from GET /v1/models or the models page.
• HTTP 404 on a path you invented. The error body lists the supported methods: chat completions, responses, completions, embeddings where the model offers them, and GET /v1/models.

The Node SDK is the same idea with baseURL in camelCase. That page is here. Pricing stays on the pricing page, because a snippet that hardcodes a rate goes stale on purpose.