Speaks first, answers fast
Click Talk and the character greets you out loud. The demo shows the measured time to first voice on every call.
Live voice character AI
Characters that talk back, live.
A first-person voice character you can simply talk to. It answers out loud in real time, lets you cut in mid-sentence, and only says what's in its brief.
Try the live demoLook around freely. Starting a call needs the password from Adi's message, and your browser will ask for the microphone.
What you'll hear
Interrupt any time
Talk over it and it stops mid-sentence, then answers what you just said. No push-to-talk, no waiting your turn.
Stays in character, stays grounded
It answers only from the knowledge you give it. Ask something off-topic and it says so in character, then steers back.
Any character in seconds
Name, persona, voice, a pasted FAQ and an opening line. Save, and you're talking to your own character.
Try this in 60 seconds
- Ask Sol which drinks it serves and what they cost.
- Interrupt Sol halfway through an answer with a new question.
- Ask something it can't know, like the weather in Paris.
- Create your own character and talk to it.
How it works
Your voice goes straight to the model
The browser streams your mic to Gemini Live over a WebSocket and plays the reply as it arrives. No audio is relayed through a middle server, which keeps it fast.
The API key stays on the server
The server only issues a short-lived, single-use token with the character's persona, knowledge and voice locked in. The API key never reaches the browser, and the token can't be reused for anything else.
Live captions on both sides
What you say and what the character says are transcribed in real time and shown under the call.
Built to grow
The same character can move beyond the browser:
- Phone numbers (Twilio / Plivo)
- Live lookups in your own data: orders, bookings, stock
- Conversation logs and analytics
- Embeddable website widget