Nuh miss a call.
Turn your AI agent into
an AI receptionist.

Not if you use the recommended setup (MCP). You paste the key into your agent's own configuration file — the same place any other app credential goes — and the agent's underlying software reads it from there to talk to expert.com.jm. The AI model itself only ever sees "book this appointment," never a key, a password field, or anything that looks like one.

We started with a simpler setup where the AI had to include the key in every request it made itself. That works, but it means the key passes through the AI's own "thinking" at some point — which is exposed to conversation logs, and to someone tricking the AI into repeating things it shouldn't. We still support that simpler setup for agents that can't do anything else, but we don't recommend it anymore now that there's a better way.

At first we just had the agent check that the name and email it was given matched what's on the booking. The problem: that only proves someone knows the client's email — not that they are the client. Anyone with that same information could get someone else's appointment cancelled.

So now, when your agent is asked to cancel a booking, nothing happens right away. We email the client a link, and the appointment stays exactly as it was until they click it. That's real proof it's actually them — not just someone repeating a name and email back to the agent.

When a client pays for a booking, going through checkout and paying is already proof that a real person made the booking. A free booking had no equivalent — it used to go straight onto the calendar the moment the agent said so, with no proof anyone real actually agreed to it.

Now a free booking works the same way as a cancellation: we hold the time and email the client a link. It only becomes a real appointment once they click to confirm it — nothing shows up on your calendar before that.

If we didn't hold the slot at all, someone else could grab that exact time in the couple of minutes it takes a client to check their email — frustrating for everyone. If we held it forever, an appointment nobody ever actually confirmed could sit there blocking that time slot indefinitely, quietly costing you real bookings.

So we hold it briefly — currently 20 minutes — and the client is told exactly how long they have. Long enough to open an email, short enough that someone who never follows through doesn't cost you the slot for good.

No, never. A scheduling mistake is easy to undo — a wrong booking or cancellation can just be redone. A money mistake isn't. If an AI agent misreads a conversation and refunds someone, that's real money gone, not a minor inconvenience. So refunds are something the agent can never do — that decision always stays with you, made from your own dashboard.

Think of it like what you'd hand a new front-desk hire on their first day: enough access to manage the calendar for your whole team, and nothing else. No pricing, no subscription or billing details, no payout or bank information. The AI agent gets exactly that same level of access — no more, no less.

No — and if you're wondering because that's happened to you elsewhere: we actually caught and fixed exactly that bug early on. Booking one barber for 2pm was making every other barber (and even the owner) look busy at 2pm too, even though nobody else's schedule had actually changed. Every team member's calendar is now tracked completely separately, so one person's bookings only ever affect their own time.

Only if you allow it. You choose how much notice you need for a cancellation — shown to clients right when they book, and you can change it any time from Settings. If someone tries to cancel through the agent with less notice than that, the request is automatically turned down and they're told how much notice you actually require. The agent won't just cancel it anyway and leave you to deal with it.

This is only about timing — it has no effect on any non-refundable booking fee you already charge for in-person services. The two are completely separate.