A voice agent that picks up when chat is not enough

What the voice layer actually does

• Interruptible, not turn-based

  Speaking over the agent cuts its output immediately, the way a real conversation works. Waiting politely for a bot to finish is the tell that you are talking to a machine.

• You can see it listening

  A canvas visualiser renders the live audio spectrum and pulse rings driven by real microphone input, so there is never a silent, ambiguous moment.

• Same brain as the chat

  Voice answers come from the same grounded engine as text, so the phone story and the chat story never contradict each other.

• Real telephony when you want it

  Add Retell, Vapi or Twilio credentials and callbacks dial for real. Without them the request is queued to the CRM and clearly labelled as queued — never a fake ring.

How it gets built

1. 1

   Build a concierge from your URL in the Train studio.

2. 2

   Open the demo and press the phone icon to start a browser voice session.

3. 3

   Add Retell, Vapi or Twilio credentials in your environment to enable real outbound calls.

4. 4

   Callbacks and transcripts land in the owner hub alongside chat leads.

Questions people actually ask

Which browsers support in-browser voice?

Chrome and Edge have the on-device speech recognition this uses. Elsewhere the agent says so plainly and offers a phone callback instead.

Do I need a telephony account to try it?

No. Browser voice works with no credentials at all. Phone callbacks need Retell, Vapi or Twilio, and without them the request is logged rather than dialled.

Is the audio sent to a third party?

In browser mode, speech recognition runs through the browser own speech service and only the resulting text reaches the concierge. Phone mode routes through whichever provider you configure.