Create an agent
Start from a template and make your first test call.
Share and embed
A share link, a one-line website widget, or an iframe.
What’s included
- A realtime AI model that follows your prompt, thinks about its answers and knows when to stop talking and listen.
- Speech recognition of the caller, in the language they speak.
- A natural voice, from a library of multilingual and English voices, with adjustable speed.
- A realistic avatar from the avatar library, or your own custom avatar.
- Natural turn-taking and interruptions, tuned per agent.
- Tools: up to 20 HTTPS endpoints the agent can call mid-conversation to look things up or take actions.
- A transcript of every call in the dashboard, and an optional signed post-call webhook.
How a call works
1
The caller opens your agent
From its share link, the widget on your website, or an iframe. Nothing to install: it runs in any modern browser, on desktop or mobile.
2
They click Start call
The browser asks for the microphone. The avatar appears, and the agent says its first message (or waits for the caller to speak first).
3
They talk
Captions show what the agent says, and a countdown shows the time left. The caller can mute, interrupt, or end the call.
4
The call ends
When the caller hangs up, the agent ends the call, or the maximum length is reached. The call, its transcript and its cost appear on the agent’s Calls tab, and your webhook receives it.
Limits
Every call is billed to the agent’s account, whoever makes it. Set a monthly spending limit, or make the agent private, to control who can call and how much it can cost.