Voice AI

A voice agent session with live transcript, intent detection and a booking tool call.

Simulated

Realtime voice

Speech-to-text
LLM
Text-to-speech

Front desk voice agentReady

Northwind Dental

AI receptionist, inbound line

00:00

Call time

Start the call

You'll hear a patient book a cleaning with the clinic's voice agent.

Speed

Live transcript

Phone numbers are redacted

No call yet

Start the call and the transcript streams in word by word, with the agent's tool calls inline.

Intent

Detected once the caller says why they're calling.

Entities

0/5 filled

Patient name: Not captured
Service: Not captured
Preferred date: Not captured
Preferred time: Not captured
Phone: Not captured

Tool calls

3 available

check_availabilityIdle
book_appointmentIdle
send_smsIdle

How it works in production

The demo above runs on a script in your browser. This is the architecture it stands in for.

Telephony

Twilio

Media streams

Realtime loop

Speech-to-text

Streaming

LLM

Intent and tool calls

Text-to-speech

Tools

check_availability

book_appointment

handoff_to_staff

After the call

Call summary

SMS confirmation

Call log

Realtime voice loop

Audio streams in both directions over the phone line. Speech is transcribed as it arrives, the model decides what to say or which tool to call, and the reply is synthesised and streamed back, with barge-in so callers can interrupt naturally.

Bookings go through the same availability API the front desk uses. When the caller asks for something outside the agent's scope, it hands off to a person with the transcript and extracted details attached, so nobody has to ask twice.

Build something like this

Tell me about the process you want to improve and the systems it touches. I'll come back with questions and a suggested approach.

Build something like this