How your agent works

What it is

Your booking agent is software. It is built on a language model, so it can understand what a guest writes and answer in ordinary language. It is not a person, and it never pretends to be.

Guests are told this in the chat window before the first message, and it stays there for the whole conversation. It cannot be switched off, whichever personality you choose.

What it does

It looks up real availability in your booking system while the guest is typing. It holds a slot, creates the booking, and sends the guest on to your booking system's own payment page. Card details never pass through the chat.

It also answers questions about you: your cancellation rules, what an experience involves, where you are. Those answers come from what you have approved.

What it does not decide

Prices, open times and group sizes come from your booking system. The agent does not work them out and does not guess them. It finds them, shows them, and uses them.

Before a booking is created, a fixed check in the code verifies all of it again: is this the right experience, is the time still free, does the price match what the guest was told. The language model takes no part in that step. It cannot invent a price or a rule, not even if a guest tries to talk it into one.

When it fetches you

It asks for a human as soon as it is not certain. That happens with a group outside your sizes, with company events and quotes, with anything that cannot be booked directly, and every time a guest asks to speak to you.

It says so plainly instead of guessing. A handoff is not a failure. It is the right exit when the answer is not certain.

Where it can be wrong

It can misread a message. It can fail to recognise the name of an experience if a guest writes it in a way it has not seen before. Then it asks again, or it fetches you.

Language models can word things wrongly. That is why times, prices and rules never come from the model, but from your booking system and from what you have approved. It is the most important boundary in the whole design, and it is worth knowing.

What it stores

Conversations are kept so you can see what happened and so the agent can improve. A guest's email or phone number is stored as a hashed reference in the decision log, not as the contact itself.

Guests get their own privacy notice in the chat. You are the controller for your guests' data, we are the processor.

How to turn it off

You can switch the agent off in your dashboard, and you can set it to answer only outside your staffed hours. You can leave it running around the clock. We do not recommend it. Take the conversations yourself when you can, and let the agent take the ones you would have lost anyway.

What we expect from you

Let the staff who watch the chat or follow up on handoffs read this page. The EU AI Act asks businesses using AI to understand the tool they are using. This page is written for that purpose.

If anything here does not match what you see in practice, tell us. It is either a fault in the agent or a fault in this text, and we would like to fix both.