Voice agent
Ask us out loud.
The agent answers.
Maya knows what we build, how we work and how long a project takes. Press and talk — she answers out loud, while you speak.
Demo
Try it now.
No sign-up, no install. You need a microphone and half a minute.
This demo holds a conversation. On the phone line the agent also books appointments, hands over to a person and sends a summary — not here.
A conversation runs up to 5 minutes. At most 3 run at the same time, so if it is busy, try again in a minute.
How we wired it in
Our side of the integration.
Verge builds the platform. This is what we wrote so the agent could live on this page — no SDK, plain WebSocket and Web Audio.
Transport
Plain WebSocket, raw PCM
No SDK, no library. We send microphone audio up, take the agent's audio down and push it into a playback queue. Text and audio travel the same connection.
Barge-in
Cut it off mid-word
The microphone never stops, not even while the agent is talking — otherwise interrupting would not exist at all. Speak up and playback stops mid-word, then the agent picks up your point.
Playback
An AudioWorklet, not chained buffers
Audio arrives in small pieces. Chained buffers leave audible gaps at the seams and an interrupt through them is never instant, so playback runs through our own worklet with a single queue.
Interface
Both themes, keyboard, phone
The panel is built on the same tokens as the rest of the site, so light and dark work on their own. The transcript is in aria-live, and the whole flow works by keyboard and on a narrow screen.
Want one like this?
The platform is Vokal.
The agent is built and run by Verge, our sister studio. We wired it into our own site and filled it with what we know about ourselves.
For your business — calls, bookings, frequent questions — talk to them. They set the agent up; we can tell you what wiring it in looks like.