Project setup & your first agent

In this chapter, you will install the LiveKit CLI, scaffold the dental receptionist project, understand every line of the generated code, run your first voice AI agent, and speak to it. By the end, you will have a working agent that responds to your voice in real time.

What you will build: A running voice AI agent that you can talk to in the LiveKit Playground.

Install the LiveKit CLI

The LiveKit CLI (lk) is your primary tool for creating, running, and deploying agents. It scaffolds projects, manages authentication, and connects to LiveKit Cloud.

brew install livekit-cli

winget install LiveKit.LiveKitCLI

curl -sSL https://get.livekit.io/cli | bash

lk cloud auth

Scaffold the project

The lk agent init command creates a complete agent project with all the boilerplate handled for you.

lk agent init dental-receptionist --template voice-agent-python

This creates a dental-receptionist directory with the following structure:

dental-receptionist/
agent.py          # Your agent's main file — this is where all the logic lives
requirements.txt  # Python dependencies (livekit-agents, plugins)
.env.example      # Template for environment variables (API keys)
Dockerfile        # For deployment to LiveKit Cloud (Chapter 12)

Move into the project directory and install dependencies:

cd dental-receptionist
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
pip install -r requirements.txt

Copy the environment template and add your API keys:

cp .env.example .env
# Edit .env and add your OPENAI_API_KEY, DEEPGRAM_API_KEY, etc.

The generated agent code

Open agent.py. Here is the complete generated file:

from livekit.agents import AgentServer, rtc_session, Agent, AgentSession

server = AgentServer()


@server.rtc_session
async def entrypoint(session: AgentSession):
  await session.start(
      agent=Agent(instructions="You are a helpful voice assistant."),
      room=session.room,
  )


if __name__ == "__main__":
  server.run()

This is a complete, working voice AI agent. Every line matters. Let us walk through each part.

Run your agent

Start the agent in development mode:

python agent.py dev

You should see output indicating the agent is running and connected to LiveKit Cloud. Dev mode does two important things: it auto-reloads when you edit your agent file, and it registers the agent with the LiveKit Playground so you can test it immediately.

Talk to your agent

Make it a dental receptionist

The default agent is generic. Let us give it a personality. Open agent.py and update the instructions:

from livekit.agents import AgentServer, rtc_session, Agent, AgentSession

server = AgentServer()


@server.rtc_session
async def entrypoint(session: AgentSession):
  await session.start(
      agent=Agent(
          instructions=(
              "You are a friendly receptionist at Bright Smile Dental clinic. "
              "Greet callers warmly and help them with appointment inquiries. "
              "Keep your responses short and conversational — one to two sentences at most."
          ),
      ),
      room=session.room,
  )


if __name__ == "__main__":
  server.run()

Save the file. Because you are running in dev mode, the agent reloads automatically. Go back to the Playground, disconnect, and reconnect to start a fresh session.

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

The agent should now respond as a dental receptionist — something like "Welcome to Bright Smile Dental! I'd be happy to help you with an appointment. What day works best for you?"

Try saying: "What are your office hours?"

The agent will answer based on the LLM's general knowledge. It might guess or make something up — that is fine for now. In Chapter 5, you will add tools that let it look up real data instead of guessing.

Try saying: "Can I speak to a human?"

Notice how it handles this. With no specific instructions about transfers, it will do its best. In Chapter 4, you will write detailed instructions that cover edge cases like this.

What you have built so far

You now have a running dental receptionist agent with these components:

• AgentServer managing the runtime
• @server.rtc_session dispatching incoming calls to your entrypoint
• AgentSession representing each live conversation
• Agent with custom instructions defining the receptionist persona
• LiveKit Playground as your testing interface

The agent can greet callers and have basic conversations, but it cannot check real data, book appointments, or handle complex workflows. In the next chapter, you will configure the models that power the voice pipeline — choosing specific STT, LLM, and TTS providers — and learn about the session lifecycle that governs every conversation.