Skip to main content
Cartesia Managed Agents helps you deploy realtime voice agents with ultra-low latency and industry-leading accuracy. Built on best-in-class speech models, it handles listening, turn-taking, interruptions, tool calls, and knowledge retrieval around the LLM of your choice. Create and test an agent in the Playground, or manage the same agents through the API. When you create an agent, the response returns its ID in the id field, such as agent_Fo7pKNBUwLZxrTd6jvhpaE. Other endpoints call this ID agent_id. Reuse this ID for new conversations instead of creating the agent again. Continue using it after restarting the browser, mobile app, or server that connects users to the agent. During a WebSocket conversation, the connected browser, mobile app, or server uses a live two-way connection to stream audio with Cartesia. For phone calls, telephony connects the agent to the phone network instead. You can give a voice agent three kinds of tools. See Tools for details. System tools are built-in actions that Cartesia handles, such as ending a call. Webhook tools send requests over the web from Cartesia to your server. Client tools are functions that the connected browser, mobile app, or server runs at Cartesia’s request.

Create your first agent

1

Create the agent

In the Playground, click Create New, then pick a template, such as Customer support, or choose Start from scratch.
Agents page in the Playground with an arrow pointing to the Create New button
2

Write the instructions

Type the System prompt and the Introduction, the first thing the agent says, directly in their fields, or click Edit with AI to draft or revise them.
Configuration page with arrows pointing to the Introduction and System prompt fields and the Edit with AI button
3

Test it

Click Preview to talk to the agent in your browser and test your changes before you publish.
Configuration page with an arrow pointing to the Preview button
4

Publish

Click Publish and review the listed changes. Phone calls and API sessions use the published version.
Configuration page with an arrow pointing to the Publish button

Explore

Agent configuration

Set instructions, language, voice, and audio behavior.

Managed LLMs

Choose an LLM and generation settings.

Tools

Add built-in actions, webhooks, and client functions.

Versions

Inspect configuration history and restore an earlier version.

Audio integration

Stream audio from an application or connect a phone number.

Evaluations

Measure call quality with built-in and custom metrics.