A phone agent that answers in about a second — and hands over without a hang-up.
Settl's voice agent hears the caller and answers in its own voice, looks up the live order, takes the action after a spoken confirmation, and — when the caller asks for a person — brings one onto the same call.
Fast enough to feel like a conversation.
Speech-to-speech means one realtime session hears the caller and answers in its own voice, with no transcribe-think-synthesise relay on the critical path. The pause after the caller stops is about a second.
It's still fully auditable: both sides are transcribed as the call happens, so the transcript, the handoff briefing and the analytics are all there afterwards.
It does things, not just talks.
The voice agent uses the same tools as every other channel. While a lookup runs, it holds the line naturally instead of leaving dead air.
- Track — the live order, rider position and ETA.
- Answer — from your policies, with the citations removed for speech.
- Act — cancel, refund, reschedule, book a test ride, after a spoken confirmation and a phone-number identity check.
- Transfer — to a person, on the same call.
“Get me a human” — without the hold music or the call-back.
When a caller asks for a person, the agent keeps them on the line, opens the conversation in your inbox with the transcript so far, and rings the agents who are available. The one who accepts joins the same call from inside the inbox; the AI steps aside and keeps transcribing. If nobody is free within a minute, the caller is told so and a ticket with the full transcript is waiting.
Every call ends up somewhere you can find it.
Calls the AI resolves are filed in the Voice inbox as resolved conversations, with a summary note — caller, topic, order, outcome, duration — and the turns as messages. They count in analytics like any other channel.
English, Hindi, Hinglish
Callers switch languages mid-sentence and the agent follows.
Two stacks, one switch
Realtime speech-to-speech by default, or a cascade with Deepgram and Sarvam voices.
Web, app and phone
Calls from your site or app today; phone numbers connect over SIP.
Confirm before acting
Spoken confirmation and identity check before any write; your backend enforces policy.
Policy or a person
Answers come from your documents; below the confidence gate, it transfers.
Calls in the dashboards
Containment, handle time and topics, sliced by voice like every other channel.
Frequently asked
What is an AI voice agent?
It's software that answers phone calls in natural speech, understands what the caller needs, and does the work — looking up an order, booking a slot, raising a refund — the way a trained agent would, handing over to a person when it should.
Which languages does it speak?
English, Hindi and Hinglish, including switching between them mid-sentence.
How fast does it respond?
We measure the time from the caller stopping to the agent starting to speak at roughly 0.9 to 1.2 seconds with speech-to-speech, against about 1.65 seconds for a classic speech-to-text, LLM, text-to-speech cascade.
Can it take refunds or cancel orders on a call?
Yes, after the caller confirms out loud and their phone number matches the order. Your order system still enforces its own rules — it won't, for example, cancel an order that's already delivered — and actions above your limits go to a person.
Can we use our existing phone number?
Yes. Browser and in-app calling are built in, and a phone number connects to the same agent over SIP through your telephony provider.