Deploy to LiveKit Cloud

Your dental receptionist works perfectly on your local machine. But a receptionist that only works when your laptop is open is not much of a receptionist. In this chapter, you will deploy your agent to LiveKit Cloud — a fully managed infrastructure that runs your agent 24/7 with automatic scaling, monitoring, and debugging tools. Six steps from local dev to production.

What happens during deployment

When you deploy to LiveKit Cloud, your agent code is packaged into a Docker container, uploaded to LiveKit's container registry, and launched on managed infrastructure. LiveKit handles scaling — spinning up more instances when call volume increases and scaling down when it drops. You do not manage servers, load balancers, or container orchestration.

Step 1: Create the agent on LiveKit Cloud

First, register your agent with LiveKit Cloud. This creates an agent entry that you will deploy code to:

lk agent create dental-receptionist

This tells LiveKit Cloud that an agent called dental-receptionist exists and will accept deployments. The name must be unique within your project.

Step 2: Write the Dockerfile

LiveKit Cloud runs your agent as a Docker container. Create a Dockerfile in the root of your project:

FROM python:3.12-slim

WORKDIR /app

COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

COPY . .

CMD ["python", "agent.py", "start"]

Make sure your requirements.txt includes all your dependencies:

livekit-agents[codecs]~=1.6
livekit-plugins-openai~=1.1
livekit-plugins-deepgram~=1.1
livekit-plugins-cartesia~=1.1
livekit-plugins-silero~=1.1

Step 3: Set secrets

Your agent needs API keys for OpenAI, Deepgram, Cartesia, and any other services it uses. Never bake these into the Docker image. Use LiveKit Cloud's secret management:

lk cloud secrets set OPENAI_API_KEY=sk-...
lk cloud secrets set DEEPGRAM_API_KEY=...
lk cloud secrets set CARTESIA_API_KEY=...

These secrets are encrypted at rest and injected as environment variables when your agent container starts. They are never visible in logs, images, or the Cloud dashboard.

Step 4: Deploy

With the Dockerfile ready and secrets configured, deploy your agent:

lk agent deploy

This command builds your Docker image, pushes it to LiveKit's container registry, and starts rolling it out. You will see progress output as the build and deployment proceed. A typical first deployment takes 2-3 minutes. Subsequent deployments are faster because Docker layer caching kicks in.

If you need to target a specific agent name (useful when you have multiple agents):

lk agent deploy dental-receptionist

Step 5: Check logs

Once deployed, check that your agent is running correctly:

lk cloud logs dental-receptionist

This streams the logs from your running agent containers. You should see startup messages confirming the agent registered successfully. When a caller connects, you will see session lifecycle events, tool calls, and any print() statements in your code.

Step 6: Test from Playground

Open the LiveKit Playground from your Cloud dashboard. Select your project, and your deployed dental-receptionist agent should appear as an available agent. Click connect and start talking — the audio now flows through LiveKit Cloud's infrastructure rather than your local dev server.

Try saying: "Hi, I'd like to book an appointment for next Wednesday."

The experience should be identical to local dev mode. If the agent responds slower, check your deployment region — deploying close to your STT/LLM/TTS providers reduces inter-service latency.

Cloud Insights: monitoring your deployed agent

LiveKit Cloud Insights gives you visibility into every conversation your agent handles. From the Cloud dashboard, you can access:

Transcripts — Full text transcripts of every conversation, showing both user and agent messages with timestamps. Invaluable for reviewing how the agent handles real calls and identifying areas for improvement.

Traces — Detailed timing breakdowns for each turn: how long STT took, how long the LLM took, how long TTS took, and the total end-to-end latency. This is where you spot performance bottlenecks.

Audio recordings — Listen to actual conversations. Reading a transcript is useful; hearing the tone, pacing, and naturalness of the interaction tells you far more about quality.

# View recent sessions
lk cloud sessions list dental-receptionist

# View details for a specific session
lk cloud sessions view <session-id>

Monitoring checklist

Once your agent is deployed, establish a monitoring routine:

Rollback if something goes wrong

If a deployment introduces a bug, roll back to the previous version:

lk agent rollback dental-receptionist

This immediately reverts to the last known-good deployment. New calls are routed to the previous version. Always test in Playground after a deployment before considering it stable.