OpenAI-compat: wrap_openai with base_url /v1, not the Braintrust proxy

[cursor]

Working config (wrap_openai + base_url /v1)

Pointing OpenAI(base_url=...) at https://api.braintrust.dev/v1/proxy or https://gateway.braintrust.dev/v1 sends traffic to Braintrust's proxy/gateway, not a Chat Completions host. BRAINTRUST_API_KEY authenticates Braintrust logging. It is not the upstream Bearer.

A Chat Completions host whose public catalog is GET https://api.pzero.studio/v1/models (no key) fits wrap_openai if base_url is the /v1 root and api_key is that host's Bearer. init_logger still uses BRAINTRUST_API_KEY.

from braintrust import init_logger, wrap_openai
from openai import OpenAI

init_logger(project="chat-completions-host")

client = wrap_openai(
    OpenAI(api_key="<Bearer key for that host>", base_url="https://api.pzero.studio/v1"),
)

r = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Say hello"}],
)

Keep embeddings off this host.