OpusDone is an AI receptionist that answers questions, qualifies enquiries and books appointments, then hands anything else straight to you. Your front desk is one person doing four jobs at once, and this takes the three that repeat. It works at 23:40 on a Sunday exactly as it does at 10:00 on a Tuesday.
Reviewed

Nothing below needs anyone at the counter. Every step is something the assistant does on its own, or hands to you the moment it should.
Someone asks what time you open and takes the first slot of the morning, with availability re-read at the moment of writing so the same slot cannot go out twice.
One is on your website and one is on Telegram, and both get the same answer from the same approved knowledge, because every channel enters the same function.
A visitor attaches a PDF quote; the assistant confirms it arrived, says plainly it cannot open it, escalates, and your inbox shows “📎 File: invoice.pdf”.
The EN/ES keyword guardrail catches the request even if the model misses it, and the escalation email goes out with who, which channel, why, three bullets, the recent transcript and one button.
The booking is still made and the lead still reaches your Google Sheet within a minute, while quiet hours hold only the phone-noisy channels until morning.
Enquiries do not arrive in office hours, and the ones that do arrive while you are with a customer.
An enquiry sent at nine at night gets a reply at nine the next morning, by which time two other businesses have already replied.
Opening hours, parking, price ranges, what to bring, whether you take walk-ins. Each one interrupts whatever is in front of you.
An enquiry that does not book leaves nothing behind — no name on a list, no number to ring on Friday when you have a cancellation to fill.
You do, and you do not have to be watching for it. The escalation email carries who it was, which channel they came from, why it escalated, a three-bullet summary, the recent transcript and one button to open the conversation.
Website widget and Telegram are live today, and a single function handles inbound from both, so a customer who asks about parking gets the same answer either way.
WhatsApp, Instagram and Messenger are not live yet.

Availability is read again at the instant the reply is composed, not fetched once at the start of the chat, so a slot taken two minutes ago is never offered.
Google Calendar or Cal.com.

A weekly digest and a monthly report showing bookings taken after hours, hours saved, how much your assistant closed out on its own and how that moved month on month.

A front desk is trusted for what it refuses to do as much as for what it handles. These are fixed behaviours, not settings you have to remember to switch on.
An EN/ES keyword guardrail catches “speak to a human” and “quiero hablar con una persona”. If the model somehow misses it, the runtime escalates anyway.
That same guardrail deliberately ignores the phrases “AI agent” and “chatbot”, so a prospect shopping for the product is not mistaken for someone demanding a person.
Working hours and closed days are checked before any time is offered, and offers are capped at six so nobody is handed a wall of slots.
Quiet hours default to 22:00–08:00 in your timezone and hold the phone-noisy channels only. The in-app bell and the email always land regardless.
The AI disclosure is added after the model writes, once per conversation, in the customer’s language, and it is not something a client instruction can remove.
Free to build and test with no card, and 14 days from going live before payment is due. The sandbox runs your real instructions and your real tools, then stops before writing anything — no leads, no alerts, no escalations.
Request access