Frequently Asked Questions

We are not going to claim we beat a travel agent — an agent has judgement and can phone ahead. What we do is mechanical: a Gemini model drafts the plan, then every activity is looked up in Google Places. Where a place matches, the plan carries that venue's real place ID, address, coordinates, opening hours and rating. Where nothing matches, we delete the place ID and rating the model produced rather than show you a number we cannot stand behind.

We do not trust the model's output — we check it afterwards. Once a draft exists, around eight deterministic passes run over it: daily pace, duplicate activities, spend against your budget, opening hours, night-time safety, overlapping times and route order. A separate lint engine can re-run those checks on a trip you edited by hand, and reports each one as clean, issues, skipped or error. Skipped is never folded into no problems found, because we could not check this and we checked this and it is fine are different sentences.

Treat it as a strong first draft, not a booking. Where a place verifies against Google Places you get its real hours and coordinates; where it does not, the rating and place ID come off rather than being guessed, and any match more than 70 km from that day's city centre is thrown out as the wrong city. We reserve nothing on your behalf, and hours and closures do go stale, so confirm anything time-critical with the venue before you build a day around it.

Every activity is looked up in Google Places after generation. A confident match gets that venue's address, coordinates, opening hours and rating attached. When there is no match, we clear the place ID, rating and review count the model produced — the suggestion stays, but with no authority signal behind it, so an activity with no rating is one we could not confirm exists. Google's own data still goes stale; email contact@travaio.co and we will look at it.

The two are handled differently on purpose. Accessibility can be checked, so a venue Google reports as step-free is kept even when its name sounds strenuous, and one reported as inaccessible is removed even when no keyword flagged it — but an absent field means no signal, never accessible. Dietary needs are only ever flagged, never used to delete anything: you cannot infer halal certification or nut-free preparation from a venue's name, and removing on a guess would delete safe restaurants while still missing unsafe ones. Gluten-free, nut-free and dairy-free have no signal to check against at all, so we report them as unchecked rather than counting them clean. Call ahead — we tell you what we could not check precisely so you know where to.

Your trips live in Postgres behind row-level security, so a trip row is readable by your own account and by anyone you explicitly invite to collaborate, and by nobody else. Traffic runs over HTTPS. Planning does mean sending your destination, dates and preferences to processors acting on our behalf: Google, for the model that drafts the plan and the Places lookups that check it, and Stripe if you subscribe. We do not sell your data or pass it to advertisers.

Your quiz answers — pace, rhythm, budget, the twelve interest scores, cuisine, and any dietary, accessibility or mobility needs you enter — plus the trip inputs themselves: destination, dates, traveller count. Then the itineraries you save and the edits you make to them, the thumbs up and down you give individual activities, your account email and display name, and your subscription state as Stripe reports it. You can download all of it from your account settings.

Yes, from your account settings. Deleting the account removes your user row, and everything keyed to it goes with it: saved itineraries, journal entries, activity comments and votes, collaborations, notification preferences, device tokens and any API keys you generated. Feedback you left is stripped of identifying details rather than deleted. Export a copy first if you want one, because deletion is permanent and we cannot restore it afterwards.

Only from things you tell it deliberately. Thumbs up and down on activities are tallied by category and carried into your next generation twice over: as a line the model reads, and as counts a separate pass acts on — one thumbs-down is a bad afternoon, five is a verdict, and it treats those differently. Places you have already been bias the next trip toward ones you have not. There is no model training on your data, and nothing learns from what you merely looked at.

It's currently in the works! Look out for updates on X and Instagram.

Saved itineraries stay readable and exportable — they are stored, not regenerated on demand. New generation depends on Google's model and Places APIs, and if either is unavailable the request fails with an error rather than handing you a degraded plan. A scheduled checker polls our dependencies every five minutes and publishes what it finds on the status page. We are a small team, so there is no staffed round-the-clock desk behind that.

A short single-city trip usually lands in about a minute. Longer ones take longer by design: the server budget is 90 seconds plus 20 more per day, capped at ten minutes, because each leg is drafted separately, short legs are re-run, and then every activity is checked against Google Places. The progress bar follows real stages rather than a timer — drafting runs to 85%, Places verification to 94%, route optimisation to 97%.

Yes. The traveller count feeds pacing, cost estimates and venue choice, and Pro adds collaboration, so people you invite can edit the same trip. Two honest limits: generation runs from one person's quiz answers, so it is not merging four sets of preferences into a compromise; and collaborative edits are last-write-wins, so two people editing the same day at the same moment can overwrite one another. Take turns on a day.

Swap an activity for a nearby alternative, add or delete one, reorder days — travel times are recomputed when you do — or regenerate a single day. Swaps and adds are Places lookups rather than fresh AI calls, so free and Pro users are not metered per trip, only bounded by a daily rate limit. Guests get two swaps and no adds per trip, because a guest cannot save a trip in the first place.

When Google Places matches the venue, we show its rating and review count as Google reports them. When there is no confident match we remove the rating rather than estimate one, so a missing rating means unchecked, not bad. We do not filter on a rating threshold, we have no review corpus of our own, and popularity is not taste — read the rating as one signal beside the description, not as a verdict.

You can regenerate a day, swap or delete activities, and reorder what is left. Regeneration drafts a fresh plan for the scope you regenerate rather than patching it, so hand-edits inside that scope are replaced — saved version history is the way back, keeping the last three versions on free and all of them on Pro. All of it needs a connection; there is no offline replanning.

We have not measured how much time this saves anyone, so we are not going to quote you a figure. What $12.99 buys is concrete: no daily generation cap, no cap on saved trips, up to five cities in one trip, collaborators, unlimited version history and share links, packing-list regeneration and the weather contingency tab. The free tier — ten generations a day, three saved trips, two cities — is meant to be enough to decide with. If you enjoy the researching part, this is not for you.

No number, because we have not measured one. What it does instead of you: ordering the cities and deciding how many days each gets, drafting every day, checking each place exists and is open when you would arrive, and sequencing stops so you are not crossing the city twice. That part takes about a minute. Reviewing the result is still your job, and you should.

On Pro there is no daily generation cap and no cap on saved trips; free is ten generations a day and three active saved trips, where archiving one frees a slot. Every itinerary gets the same treatment — day-by-day plan, Places-checked detail wherever a venue matches, route ordering, map, and export to PDF, ICS and KML plus the Pro-only formats. One ceiling sits behind the word unlimited: day regeneration is capped at 200 a day per account, roughly fourteen two-week trips regenerated day by day, and it exists to stop scripted abuse rather than to meter you.

A little, through two narrow mechanisms rather than a model quietly learning you. Your thumbs up and down are tallied by category and shape the next generation, and places you have already visited bias it toward ones you have not. The larger lever is still the quiz: the more precisely you set pace, interests and no-gos, the less there is left to guess at.

Regenerate the day, swap the item out, or restore an earlier version from history — free keeps the last three, Pro keeps all of them. If the trip is wrong rather than just not to your taste — a place that does not exist, hours that are impossible, a day that cannot physically be walked — email contact@travaio.co with the trip and we will look at the generation itself, not just the symptom.

Contact us via email at contact@travaio.co anytime. We typically respond within 24 hours.

The app is, and saved trips are readable whenever you are. New generation depends on Google's APIs and on our own service, so an outage on either side stops new plans while the ones you already have stay intact — live dependency status is on the status page. Support is not around the clock: it is email, answered in business hours, usually within a day.