Use the OpenAI Python SDK with a custom Base URL

What was tested?

We installed openai==2.48.0 in a clean Python virtual environment and created OpenAI(api_key=..., base_url="https://monkeytheme.com/v1", max_retries=0, timeout=20). Calling client.models.list() with an intentionally invalid key reached MonkeyAI and returned HTTP 401 with INVALID_API_KEY. That proves the SDK, TLS route and custom Base URL reached the authentication layer. It does not prove that a real key, a particular model or a generation endpoint will succeed.

How do I install and configure it?

python3 -m venv .venv
source .venv/bin/activate
python -m pip install openai

export MONKEYAI_API_KEY="***"

Do not commit the key to a Python file. The OpenAI SDK README recommends environment-based secret loading, and the constructor accepts an explicit api_key value.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["MONKEYAI_API_KEY"],
    base_url="https://monkeytheme.com/v1",
)

How do I verify the key before generating text?

models = client.models.list()
for model in models.data:
    print(model.id)

A successful list gives you the model IDs visible to that key. Copy an exact ID into later requests instead of guessing. A 401 means the request reached the service but authentication failed. If the call fails before receiving an HTTP response, inspect the network or TLS exception instead.

How should I handle errors?

import openai

try:
    models = client.models.list()
except openai.AuthenticationError:
    print("Check MONKEYAI_API_KEY")
except openai.APIConnectionError as exc:
    print("Network or TLS failure:", exc.__cause__)
except openai.APIStatusError as exc:
    print("HTTP status:", exc.status_code)
    print("Request ID:", exc.request_id)

The SDK documents APIConnectionError for connection failures and APIStatusError for non-success HTTP responses. It also exposes a request ID on status errors when one is available. Log the request ID, not the API key.

What about retries and timeouts?

The official SDK README says selected connection errors, 408, 409, 429 and server errors are retried twice by default. It also documents a ten-minute default request timeout. Those defaults can hide repeated work during debugging, so the verification example uses max_retries=0 and timeout=20.

client = OpenAI(
    api_key=os.environ["MONKEYAI_API_KEY"],
    base_url="https://monkeytheme.com/v1",
    max_retries=0,
    timeout=20.0,
)

Choose production retry and timeout values for your workload. A short model-list check and a long streaming generation should not automatically share the same policy.

Which generation API should I call?

The OpenAI SDK supports both responses.create() and chat.completions.create(). An OpenAI-compatible relay may implement a different subset. Check the MonkeyAI guide for your client or endpoint before choosing one. We did not run authenticated generation in this verification, so this page deliberately stops at model discovery.

Sources

• OpenAI Python SDK README, pinned revision
• Official OpenAI Python SDK repository
• MonkeyAI API key troubleshooting

FAQ

Related guides