Step 1: Open the agent creation page

From the dashboard, go to Voice Agents → New Agent.

Step 2: Configure the basics

Name — internal label to identify this agent
Instructions (system prompt) — tell the agent what its role is, what tone to use, and any constraints. Example:

You are a friendly receptionist for Acme Dental. Help callers book appointments, answer questions about insurance, and transfer to a human if asked. Keep responses under 2 sentences.
Welcome message — the first thing the agent says when a call connects

Step 3: Pick your stack

Each agent uses three AI models:

Component	What it does	Recommended default
LLM	Generates responses	OpenAI GPT-5 mini (via OpenRouter)
STT	Converts speech to text	Deepgram Nova-3
TTS	Converts text to speech	OpenAI TTS (Alloy voice)

You can swap any of these at any time — agent configurations live in the database and are read by the LiveKit agent at call time.

Step 4: (Optional) Configure failover

For production agents, enable Failover Backup on each stack component. If your primary provider has an outage, the agent automatically retries with the backup mid-call — no dropped calls.

Common pairings:

LLM: OpenAI GPT-5 mini → backup Claude Haiku
TTS: OpenAI → backup Cartesia
STT: Deepgram → backup AssemblyAI

Step 5: Save and test

Click Create Agent. You'll be redirected to the agent detail page where you can:

Test the agent in your browser (no phone number needed)
Connect a phone number for inbound calls
Review call history after each session

Next steps

Set up a phone number →
Configure integrations (Google Sheets, Calendar)

Create Your First Agent