Everything your customers type, answered.

A website visitor becomes a booking, not a callback

The same booking engine your phone line uses.

When a visitor is ready, the chat checks your live calendar inside the conversation, offers what is open, and books it. No form, no promise to call back, no phone tag with somebody who was ready to buy ten minutes ago.

It runs the same booking code the phone receptionist runs, with the same honesty rule: it only says booked when a real calendar event came back. If it cannot reach your calendar it says so and takes the details instead, rather than promising a slot it could not verify.

Mentions the upgrade you actually sell

From a list you write. It cannot invent one.

When somebody is close to booking and it fits what they came for, the chat can mention one upgrade: the maintenance plan, the bigger package, whatever you offer. You write that list in your dashboard, up to eight, in your own words.

It can only name something on that list. It will not invent a price, a term or an inclusion, and if someone asks for a detail it does not have, it says you will confirm rather than making a number up. It stays quiet on a first message about an urgent problem, with anyone who is upset, and if they ignore it or say no it drops the subject for good. Leave the list empty and it never upsells at all.

Talks about your business and nothing else

Ask it to write your homework and it politely declines.

A chat widget on a public website is a free assistant for anyone who finds it, and plenty of people will try. This one stays on your work: ask it to write code, do homework or opine on the news and it declines and steers back.

The harder problem is people who type instructions at it trying to make it ignore its own. Every piece of untrusted text reaching the model, including what a visitor types and what a reviewer wrote, is wrapped so it reads as data rather than orders. Worth being straight about this: that is a mitigation, not a proof. It is paired with keeping the assistant unable to do much damage even if it is talked around, which is why it holds so few tools and why nothing that sends in your name goes out without your rules being met.

Will answer every review without ever asking anyone to change one

Built and waiting on Google. Here is exactly what it will and will not do.

This half is not switched on. Reading your reviews and posting a reply needs Google Business Profile access, Google grants that on their own schedule, and we will not put a date on somebody else’s approval queue. Facebook has no posting path at all yet. None of it changes what a tier costs.

What it does the day it opens: every Google review gets a reply written from your business, your services, your hours, your tone. A replied-to review tells the next person reading it that somebody is paying attention, and that person is the one deciding whether to call you.

The negative ones follow a strict rule. It apologises, takes the conversation offline, and never asks the reviewer to update, remove or reconsider what they wrote. That last part is not politeness, it is the law: pressuring people over reviews is what the FTC goes after. It also never invents a fact, a price or an offer that was not in what you told us.

One tap, or hands-off, and a bad review will never send itself

The rules that will govern review replies the day they open. Auto-send is yours to switch on, and it is fenced.

By default every reply is a draft. You read it, you tap once, it goes. Nothing is ever sent in your name that you have not seen.

Turn auto-send on and three things must all be true before anything leaves: you switched it on, the draft is not negative, and the model cleared the confidence bar you set. A negative draft is never auto-sent, however confident it is, because the reply that most needs a human is the one where somebody is upset. The routine thank-yous clear themselves and everything with a knot in it waits for you.