Skip to content
Documentation
GuidesAPI ReferenceIntegrations
ArenaPlatform
Integrations/Voice Frameworks

LiveKit

Use AudioLab as the language layer of a realtime agent.

Prerequisites

Complete the integration setup and verify a successful text request before configuring this client. Review the shared compatibility checks alongside the client-specific limitations below.

Install the Integration

Start from an existing working LiveKit Agents application with its LiveKit credentials, transport, and audio pipeline configured. Add the OpenAI plugin in that project:

Terminal
uv add "livekit-agents[openai]~=1.5"

Connect the LLM

LiveKit’s OpenAI-compatible plugin uses Chat Completions for custom gateways. The snippet below configures only the LLM component; incorporate it into your existing application instead of running it as a standalone agent. Add your LiveKit transport, STT, TTS, VAD, and turn detector separately; the same provider credential does not make the speech plugins interchangeable.

Language Component
import os
from livekit.agents import AgentSession
from livekit.plugins import openai

session = AgentSession(
    llm=openai.LLM(
        model="openai/gpt-5.5",  # Use an enabled, tool-capable text model.
        base_url=os.environ["AUDIOLAB_BASE_URL"],
        api_key=os.environ["AUDIOLAB_API_KEY"],
    ),
    # Supply STT, TTS, VAD and turn detection for your pipeline.
)

LiveKit Compatible LLMs

Test the Full Conversation

First verify a spoken question produces a text response from the AudioLab language model and audible output through your existing TTS component. Then test interruptions, silence, disconnects, and concurrent sessions. Inspect usage after cancellation as well as after success. AudioLab’s speech WebSockets need an adapter to the framework’s audio frames and events; do not assume a base URL change implements that adapter.