Products
What your AI employee actually does
The front page says what it does — this page shows how. Every capability below is in the product today, described the way it really works, because the mechanism is the promise.
Trained on your business, on your number, on the record.
On the line around the clock
The whole job, not a phone tree
Eight things a good hire covers, and the mechanism that does each one here.
Answers 24/7
Every call to your number is picked up by your agent — nights, weekends, and the second call that arrives while the first is still talking. It is trained on your business automatically: paste your website and the Brain reads it into working knowledge of your services, your hours, and the words your callers actually use. You choose its voice from a catalogue of thousands, and audition any of them speaking your own greeting before it ever takes a call.
Takes bookings
Booking lives where it can be guaranteed: your booking page. It reads your real service menu and availability — the same documents you edit in the console — and offers only times that are actually free. A slot that is already taken is refused with the day’s other free times, never silently double-booked, and every booking made there comes with a calendar file. People who book and people who call fold into the same customer record, so one person is one history.
Qualifies leads
The agent is compiled for your industry, so it knows “knocked out” is urgent at a dental practice, “poison” is urgent at a veterinary clinic, and “arrested” is urgent at a law firm. It captures who called, why, and how urgent — and your customer list sorts the people who need a callback to the top, so the order of your follow-ups is already decided when you open it.
Takes messages
When the right answer is “I’ll pass that on”, nothing rides on anyone’s memory: the whole conversation lands in your console as a transcript with its recording, tied to the person who called. Nothing lives on a sticky note, and nothing is lost between shifts.
Knows who is calling
Import your phone’s contacts — a .vcf straight off an iPhone or Android, or a CSV — and they become the names on your customer list. Mid-call, the agent checks whether the caller is one of them, so a regular is greeted like a regular, with your notes in front of the AI while it talks. And when it cannot be sure who is on the line, it says nothing: a wrong name spoken to a caller is the one failure the lookup is built never to produce.
Transfers urgent calls
Some calls need you, and the agent knows which. What counts as urgent is a rule you set — after hours, named emergencies, a caller who asks for a human — not a guess the AI makes. When a call qualifies, it is handed to your phone instead of becoming a message about something that could not wait.
Follows up after the call
Hanging up is not the end of the job. Every finished call is tagged with how it ended and the topics it covered, and the follow-up steps you draw in your call flow’s post-call lane run after the caller is gone — a step that fails is recorded in words you can read, not forgotten. Your other systems can hear about every call too, through signed webhooks.
Syncs your tools
Signed webhooks push call events to any endpoint you give us — retried with backoff for about a day and a half, and an endpoint that keeps failing is switched off and surfaced, so a dead integration can never sit there looking healthy. There is a full read API for your balance, agents, numbers and calls — and an MCP server, so your own AI tools can ask.
What happens on every call
An AI that speaks in your name has to be honest about being one, lawful about recording, and accountable for every second it bills. None of that is optional here, in any country we sell a number.
“Kia ora, you're speaking with an AI assistant. This call may be recorded.”
The first words of every call on a New Zealand number — the wording is the country’s, not a global default.
It says it is an AI, first
Disclosure is on by default and worded per country — it is the first thing every caller hears. Regulation points one way: the EU AI Act’s transparency article applies from August 2026, and the FCC already treats an AI voice as “artificial” under the TCPA. We build disclosure in as the floor, not a setting to find.
Recording follows local law
Whether a call may be recorded follows the rules of the number’s country, from the same catalogue that decides the disclosure line. An Australian number opens with “G'day, you're speaking with an AI assistant, and this call is being recorded.” while a US number opens with “Hi, you're speaking with an AI assistant. This call may be recorded.” — Australia’s consent rule is stricter, so the wording is. The difference is the law’s, not something somebody remembered to configure.
Everything is on the record
Every finished call lands in your console with its transcript, its recording, and how it ended. What the AI said in your name is never a mystery.
Metered per second, on a ledger you can read
Credits are reserved the moment a call starts, metered per second while it runs, and settled at hang-up — you are charged for the seconds that actually happened and the remainder is released. Usage draws on your plan’s included minutes first, then your credits, and every entry is on a statement you can read.
The console behind it
Everything the agent does is visible, replayable, and yours to change — from one screen, not a support ticket.
Agent design
Voice, guidelines, knowledge, and a call flow you can see — a canvas of the steps your agent takes, not a wall of settings. Publishing is the only way changes reach the phone: what the canvas shows is what the line does, with no second copy to drift.
Conversations
The full call history — transcript, recording, and how each call ended — searchable, and tied to the person who called rather than to a bare phone number.
Live monitor
Which of your lines are in use right now, against your plan’s calls-at-once cap — “3 of 5 lines in use”, amber at the limit. It shows presence, not live captions: we display what we can actually stream, and nothing we cannot.
People
A customer list nobody had to type. Everyone who has called or booked is folded into one record with a full timeline, and the people waiting on a callback surface first. Import your contacts to put names on it — and to let the agent greet them by name.
A/B testing
Switchback tests for your agent: one variant live at a time in alternating windows, every call attributed to the variant that was live when it started. The screen refuses to name a winner until each side has enough calls to mean something — a verdict we could not compute honestly is worse than none.
Hear it answer for your business
Paste your website and talk to your own agent in the browser — before you pay for anything.