Check-in instructions only for real guests: Celeste for vacation rentals

Guest questions cluster at the two moments a host is least available: the week before arrival, when excitement turns into logistics, and the first hour on the property, when the thermostat wins. Is there parking for two cars. Do you have a crib. What is the wifi password. How does the hot tub work. What time can we check in, and can we check in early. Hosts and small rental operators answer these from phones, at dinner, on the highway, because the alternative is a guest standing in a driveway.

Celeste answers them on email and web chat, from the house manual you already wrote.

The house manual finally gets read

Every rental has the binder: appliance quirks, parking rules, trash day, the trick to the grill, the good taco place. Guests do not read binders; they ask. Ingest the manual, the listing details, and your policies once, per property, and the agent answers what they actually say, at 9pm, in the driveway, on the third ask of the same question. Multi-property operators keep knowledge separated per listing, so the hot tub instructions for the lake house never end up at the condo.

When a question is not in the manual, the agent says so and routes it to you rather than improvising, and the miss is logged as a named knowledge gap. Hosts learn quickly that the gap list is a to-do list: each entry is a paragraph you write once and never answer by thumb again.

The door code is not a FAQ

Check-in instructions are different from parking questions, and the difference is enforcement. Door codes, lockbox locations, and alarm details should release only to the person with the reservation, not to whoever emails asking for them. In Celeste, that material sits behind identity verification: a guest confirms who they are first, and what the agent releases is scoped to that guest’s own booking by code that runs outside the AI. Someone fishing for access details without a matching reservation gets nothing, no matter how convincingly they ask, because the lookup itself cannot return another booking’s data.

The same scoping covers reservation questions. Dates, balance, what is included: each verified guest sees their own booking and nobody else’s.

What escalates to you

Some messages need the host, immediately. Put “locked out,” “no hot water,” “flooding,” “can’t get in” on the escalation keyword list, and a message containing one is barred from an autonomous reply and held for you with the trigger flagged, so the guest in the driveway is never answered by a cheerful FAQ. Refunds, date changes, and damage claims follow the write rule that governs the whole product: the agent drafts, and it executes only what you have explicitly classified as safe to automate. A refund proposal arrives pre-written for your one-click approval, not pre-sent.

Turn it up per property

The agent practices in shadow mode against your real guest messages while you read the drafts. Amenity and house-manual questions typically graduate first. Verified check-in release goes live when you have watched the identity flow run clean. Cancellations and money stay held for as long as you want.

The hours this returns are the ones that made hosting feel like a second job: the evening interruptions, the repeated wifi password, the check-in message sent from a restaurant. The guest experience improves for the same reason, since the answer arrives at the moment of the question instead of when your phone comes back out.

See how it works or the product. The trial starts silent, so the agent reads a few weeks of your real guest questions before it ever answers one.