LiveKit Voice AI

Build voice AI agents with LiveKit. Use when developing realtime voice applications, voice agent pipelines (STT-LLM-TTS), WebRTC communication, or deploying conversational AI. Covers LiveKit Agents SDK, provider selection (Deepgram, OpenAI, ElevenLabs, Cartesia), cloud vs self-hosted deployment.

LiveKit Voice AI Skill

Build production voice agents with LiveKit's open-source platform.

Quick Start

# Install SDK
pip install livekit-agents livekit-plugins-openai livekit-plugins-deepgram livekit-plugins-cartesia

# Or Node.js
npm install @livekit/agents @livekit/agents-plugin-openai

Minimal Agent (Python)

from livekit.agents import AgentSession, JobContext, WorkerOptions, cli
from livekit.plugins import deepgram, openai, cartesia

async def entrypoint(ctx: JobContext):
    await ctx.connect()
    
    session = AgentSession(
        stt=deepgram.STT(),
        llm=openai.LLM(model="gpt-4.1-mini"),
        tts=cartesia.TTS(),
    )
    
    session.start(ctx.room)
    await session.say("Hello! How can I help?")

if __name__ == "__main__":
    cli.run_app(WorkerOptions(entrypoint_fnc=entrypoint))

Provider Selection

Component	Budget	Quality	Low Latency
STT	Deepgram Nova-3	AssemblyAI	Deepgram Keychain
LLM	GPT-4.1 mini	Claude Sonnet	GPT-4.1 mini
TTS	Cartesia Sonic-3	ElevenLabs	Cartesia Sonic-3

Voice Pipeline vs Realtime

STT-LLM-TTS Pipeline:

• More control, mix providers
• Generally cheaper
• Easier to debug

OpenAI Realtime API:

• Speech-to-speech, more expressive
• Higher cost (~$0.10/min)
• Less control

Environment Variables

LIVEKIT_URL=wss://your-app.livekit.cloud
LIVEKIT_API_KEY=your-api-key
LIVEKIT_API_SECRET=your-api-secret

# Provider keys (if not using LiveKit Inference)
OPENAI_API_KEY=
DEEPGRAM_API_KEY=
CARTESIA_API_KEY=
ELEVENLABS_API_KEY=

Tool Use

from livekit.agents import function_tool

@function_tool()
async def get_weather(location: str) -> str:
    """Get current weather for a location."""
    # Your implementation
    return f"Weather in {location}: 72°F, sunny"

session = AgentSession(
    stt=deepgram.STT(),
    llm=openai.LLM(),
    tts=cartesia.TTS(),
    tools=[get_weather],
)

Telephony (SIP)

from livekit import api

# Outbound call
await lk_api.sip.create_sip_participant(
    api.CreateSIPParticipantRequest(
        sip_trunk_id="trunk-id",
        sip_call_to="[phone]",
        room_name="my-room",
    )
)

Deployment

LiveKit Cloud: livekit-server-cli deploy --project my-project

Self-hosted:

docker run -d \
  -p 7880:7880 -p 7881:7881 -p 7882:7882/udp \
  -e LIVEKIT_KEYS="api-key: api-secret" \
  livekit/livekit-server

Cost Estimates

Scenario	Monthly Cost
Dev/testing	Free tier
100 hrs/mo voice	~$150-250
Production B2B	~$300-500
High volume	Self-host

Common Patterns

Turn Detection

session = AgentSession(
    turn_detection=openai.TurnDetection(
        threshold=0.5,
        silence_duration_ms=500,
    ),
    ...
)

Interruption Handling

@session.on("user_speech_started")
async def handle_interruption():
    session.stop_speaking()

Multi-Agent Handoff

await session.transfer_to(specialist_agent)

References

• Docs: https://docs.livekit.io/agents/
• Examples: https://github.com/livekit/agents/tree/main/examples
• Playground: https://agents-playground.livekit.io