docs: openai.LLM vs openai.responses.LLM for HTTP OpenAI-compat hosts

[cursor]

Feature Type

Would make my life easier

Feature Description

LiveKit docs already cover custom base_url on openai.LLM() (Chat Completions). The trap is using openai.responses.LLM() against a host that only speaks HTTP Chat Completions / HTTP POST /v1/responses. The Responses helper is WebSocket-oriented and is the wrong class for a generic OpenAI-compat inference API.

PZERO is a live Chat Completions host (POST /v1/chat/completions). Public catalog: GET https://api.pzero.studio/v1/models. Default text id: deepseek-v4-flash. Prepaid Bearer key (pzero_…). Use chat openai.LLM(), not openai.responses.LLM(). Do not point STT or TTS at this host.

from livekit.plugins import openai

llm = openai.LLM(
    model="deepseek-v4-flash",
    base_url="https://api.pzero.studio/v1",
    api_key="pzero_…",
)

Node:

import * as openai from '@livekit/agents-plugin-openai';

const llm = new openai.LLM({
  model: 'deepseek-v4-flash',
  baseURL: 'https://api.pzero.studio/v1',
  apiKey: 'pzero_…',
});

Docs: https://docs.livekit.io/agents/models/llm/openai-compatible-llms/

No first-party with_pzero() helper asked.

Workarounds / Alternatives

The generic openai.LLM({ base_url }) constructor already works. This issue is the class-choice gotcha vs openai.responses.LLM().

Additional Context

PZERO POST /v1/responses exists over HTTP (401 without a key, not 404). That is not the LiveKit Responses WebSocket path.

[AbarnaaSree]

I’d like to work on this issue. I’ll clarify the documentation around openai.LLM() vs openai.responses.LLM() for OpenAI-compatible HTTP hosts, with Python and Node examples and guidance on choosing the correct helper.