OpenAI-compat: AsyncOpenAI base_url /v1 then instrument_openai

Working config (AsyncOpenAI base_url /v1 + cl.instrument_openai())

Bare AsyncOpenAI() hits api.openai.com. The app-public-URL env is the UI origin, not the Chat Completions host. Enabling LangChain or LlamaIndex instrumentation at the same time duplicates steps.

A Chat Completions host whose public catalog is GET https://api.pzero.studio/v1/models (no key) fits if base_url is the /v1 root and cl.instrument_openai() runs after the client is created. Override default gpt-3.5-turbo.

from openai import AsyncOpenAI
import chainlit as cl

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

settings = {"model": "deepseek-v4-flash", "temperature": 0}

Keep embeddings off this host.