Create and Deploy Voice Agents for Your Business Using Cartesia
Voice calls still matter for businesses that take orders, handle bookings, or answer customer questions.
Many teams want automation here, but most tools feel robotic or hard to set up.
That gap is exactly where AI voice agents come in.
This tutorial explains how to create and deploy AI voice agents using Cartesia. The process covers building an agent, testing it, and publishing a live phone number that customers can actually call.
Every step uses Cartesia’s own interface and features, without external tools or workarounds.
What you will learn in this guide:
-
How to access Cartesia and test voices using Sonic 3.0
-
How to create a voice agent using Text to Agent
-
How to test the agent through a real dialer number
-
How to deploy the agent and monitor live call metrics
The walkthrough sticks to practical steps you can follow immediately.

Step 1: Access Cartesia and Test Voices with Sonic 3.0
Cartesia is accessed through its Sonic interface, which includes voice testing and agent creation in one dashboard.
The first task is to explore available voices and confirm latency and clarity before building an agent.
Follow these steps to get started:
-
Click Start for Free and open the dashboard
-
Open the Text to Speech section
-
Select Sonic 3.0 as the voice model
-
Test different voices and listen for response speed and clarity
Sonic 3.0 is recommended because it offers lower latency and clearer output compared to older models.
Testing voices first helps avoid rebuilding agents later due to poor audio quality or pacing.
Key things to evaluate during voice testing:
-
Natural pacing when reading full sentences
-
Pronunciation of numbers and names
-
Consistency across repeated prompts
Once a suitable voice is identified, it can be reused during agent creation without repeating this step.
Step 2: Create a Voice Agent Using Text to Agent
After confirming a voice, the next step is to build the actual voice agent.
Cartesia uses a Text to Agent workflow that converts written instructions into a callable AI agent.
Here is the exact process:
-
Scroll to the Voice Agents section
-
Click Text to Agent
-
Describe the agent’s role in plain language
-
Select the previously tested voice
-
Click Generate to create the agent
A sample agent description used in Cartesia looks like this:
-
Pizza order assistant
-
Greets customers
-
Takes orders
-
Confirms details
-
Calculates totals
Clear descriptions produce better agent behavior. Listing tasks in full sentences helps the system understand call flow and responsibilities.
After generation, the agent becomes immediately testable. At this stage, no production number is assigned yet, which keeps testing isolated from real customers.
Step 3: Test the Voice Agent Using the Live Dialer
Once the agent is generated, Cartesia provides a live testing option through a phone dialer.
This step verifies how the agent behaves during real calls before any public deployment.
Testing is done using the built-in dialer number:
-
+1 (515) 800-8360
Call the number and interact with the agent as a real customer would. This stage focuses on practical performance rather than setup correctness.
During testing, confirm the following points:
-
Response speed during back-and-forth conversation
-
Accuracy when taking orders or capturing details
-
Clarity and consistency of the selected voice
-
Proper handling of confirmations and totals
Live testing helps surface issues that do not appear in text-based previews. Repeating common customer scenarios gives a clearer picture of how the agent will perform in production.
If problems appear, return to the agent description and adjust the prompt.
Small changes in phrasing often improve how the agent responds during calls.
Step 4: Deploy the Agent and Monitor Call Performance
After successful testing, the agent can be deployed for real use. Cartesia handles deployment directly from the dashboard without extra configuration steps.
Deployment follows a simple flow:
-
Click Promote to Production
-
Receive a working phone number
-
Publish the number on your website or marketing materials
-
Monitor performance inside the Metrics section
The Metrics view tracks how the agent performs after going live. This includes:
-
Number of calls handled
-
Call duration
-
Credit usage
Monitoring these metrics helps determine whether the agent meets business needs or requires further refinement. Changes can be made over time as customer interactions reveal new patterns or issues.
A practical improvement loop is encouraged. Review call data, refine prompts, and retest as customer behavior evolves.
FAQs
What can AI voice agents do for a business?
They can handle calls, take orders, and answer customer questions using natural, human-like speech.
Where do you go to start building voice agents with Cartesia?
You start by visiting https://cartesia.ai/sonic and opening the Cartesia dashboard.
Which Cartesia feature allows you to test voices before creating an agent?
Voices are tested using the Text to Speech section inside the Cartesia dashboard.
Which voice model is recommended for best latency and clarity?
Sonic 3.0 is recommended for best latency and clarity.
How do you begin creating a voice agent in Cartesia?
You scroll to the Voice Agents section and click Text to Agent.
How should a voice agent be described during creation?
The agent should be described in plain language that explains its role and tasks, such as greeting customers, taking orders, confirming details, and calculating totals.
How do you generate a voice agent after writing its description?
You select a voice preference and click Generate.
How can you test a generated voice agent before deployment?
You test the agent by calling the Cartesia dialer number and interacting with it as a customer.
What phone number is used to test Cartesia’s voice agents?
The test number is +1 (515) 800-8360.
What should be verified when testing a voice agent by phone?
You should verify response speed, order handling, and voice clarity.
How do you deploy a voice agent for real customer use?
You deploy the agent by clicking Promote to Production.
What happens after promoting a voice agent to production?
You receive a working phone number that can be published and used by customers.
Where can you monitor how a deployed voice agent performs?
Performance is monitored in the Metrics section of the Cartesia dashboard.
What data is available in the Metrics section?
It shows calls handled, call duration, and credit usage.
How should voice agent prompts be improved over time?
Prompts should be refined based on customer interactions.
