Technical draft
What the artificial intelligence of this product actually does, what the providers do, and what is not there behind it. If you have two minutes, read this before you trust it with your trip.
Last updated: October 20, 2026
Artificial intelligence is the engine that turns what you write into an itinerary. Concretely, it does three things: it decides the structure of the trip (which days, in which city, at what pace and in what order), it chooses the activities and accommodation that fit your dates and budget, and it writes the narrative that explains each day.
From there the trip is yours and it is edited in the chat: if the hotel does not work, if you want one more day at the beach or if the pace is too fast, you say so and the AI rebuilds the plan. The chat is the trip editor, not a support chat.
What the AI does not do: it does not check that a place exists on a given day, it does not check the price at the moment you book, it does not know your entry requirements, and it is not right without the possibility of being wrong. It is a language model working with third-party data: it can fall short, it can mix data up and it can be wrong in a convincing way.
The AI does not invent prices, schedules or availability: it asks the providers who have them, and that is what makes it useful. But knowing where each data point comes from matters as much as the data point itself, so every element of the itinerary shows its origin and the date of the query.
The inventory providers are the source of flights, hotels, experiences and transfers. Every price you see comes from a specific query, with its timestamp, and the amount is in its currency and its unit (per person, per group or total). Opening hours and availability come from there too, not from the AI.
The weather for each day is a meteorological estimate for that date and city. Routes and transfers come from the maps and transport services. And the price history we keep records what was checked and when, which is what lets us tell you that a price is two days old rather than today's.
| Trip data | Where it comes from | What it means |
|---|---|---|
| Price of a flight, hotel, experience or transfer | The inventory provider, at the moment of the query | It is the price the provider returned, with its date; it is not booked and not guaranteed |
| Opening hours and availability | The inventory provider | What the provider states; it can change without notice |
| Weather for the day | The weather service | An estimate for that date and place, not a guaranteed condition |
| Route and transfers between cities | Transport providers and maps services | An estimate of duration and distance |
| Trip narrative and day title | The language model | Written from the data above; it can contain errors |
Other products in this sector make promises similar to ours and back them with a team of people: an advisor reviews every detail, calls the hotel, negotiates and books. Here there is no such team. We have no named advisors and we are not going to pretend otherwise.
Lyra is an automatic product. When you ask for a trip, an AI generates it with provider data and hands it over. There is no person reviewing your itinerary, no person booking for you and no support line with an advisor on the other end. If there ever were, this page would say so on the same day, and it would say so with names.
There is no margin and no intermediaries either: we do not charge commission on the bookings you make, because we do not make bookings. And the full package price is not published yet: when it is, it will still be an estimate, not a fixed-price offer.
There are destinations we have good data for and others we do not. When we look for flights, hotels or activities in a city and find nothing verified, the engine has two options: propose nothing, or propose something reasonable without checking it. So that a plan is not left half-built, the engine chooses the second, and makes it clear.
The activities in those cities are generic, unverified suggestions: places and experiences the model considers reasonable for that kind of place, with no checked price, no checked schedule and no confirmation that they still exist. The itinerary itself says so, and the price is shown as indicative.
When the data comes from a provider, the price and the schedule are the provider's, with its date. When it is an unchecked suggestion, there is no date and no fixed price, and it says so. You can see that difference on every element of the trip.
If something in the trip is wrong, the quickest route is the chat: it is the same conversation you created it with, and the AI rebuilds the plan with the change. You can ask for a hotel in another category, to drop a day, to change the pace or to move an activity to another date, and see the result straight away.
If what is wrong is data that comes from a provider (a price, a schedule), the chat cannot fix it: you have to go to the source. Every activity that comes from a provider carries its link, and that is where the information is confirmed and the booking is made.
And if you find a bug in the product, rather than in your trip, you can report it. We read it, we look at it and, when the error comes from provider data, we pass it on to whoever publishes it.