Our approach
Understanding the place before speaking about it.
A voice that tells a story is only worth hearing if it knows what it is talking about. Here is how Thaïa knows.
An answer is only useful if it understands where, when and why it is being asked.
From source to story
Five steps between a data point and a sentence.
Sources
Open corpora anyone can check — Wikidata, Wikipedia, public geographic data — plus whatever an institution entrusts to us.
Knowledge
Sources are cross-checked and linked: a place, its coordinates, its periods, its people. What does not cross-check is not kept.
Context
Where you are, what you are looking at, the time of day, what moved you earlier. Without that context, an answer can be accurate and beside the point.
Interpretation
The model writes from that knowledge base, not from its training memory. That is what lets a sentence be traced back to a source.
Experience
The voice, the pacing, the timing. An explanation that arrives too late or too loud never happened.
The engine
Five capabilities, one knowledge base.
Visual recognition
What you frame goes out with your position, the time of day and what you have already looked at. Without that context, a model describes a stone; with it, it identifies a place.
Computer vision + open geographic data.
Real-time voice
The answer starts before you finish your sentence. The stream runs both ways: you can cut in mid-sentence, the way you would with a person.
Real-time neural voice. Your voice isn't recorded.
Itinerary composition
One sentence is enough to set the constraints — duration, mood, crowds, distance. The rest is a scheduling problem, not a matter of taste.
Open sources, algorithmic curation.
Proximity
The phone watches zones, not your route. A place only triggers an alert if it clears a notability threshold — silence is the default.
On-device geofencing. No route is sent.
Memory
What is kept is a list you can read. Three months later, in another city, it serves as context — or you erase it.
You control your memory. Erasable anytime.
Illustrative previews — the app is coming soon.
Sourcing and verification
An answer must lead back to its source.
A story you cannot trace back to its source is not information, it is an impression. When an answer rests on a source you can check, Thaïa cites it. When it cannot, Thaïa says so rather than filling the gap.
Traceable
Facts come from the knowledge base, not from the model's memory.
Cited
The source is shown, and clickable when it is public.
Correctable
A reported error is fixed at the source, for everyone.
Adaptation
The same story does not fit every situation.
An eight-year-old, a specialist, someone with ten minutes and someone who cannot hear do not want the same story about the same place. That is not four pieces of content to write: it is one knowledge base and a set of constraints on how it is delivered.
Explainable and controllable
Technology must stay in its place.
Minimisation
Visit data serves the visit. What is not necessary is not collected.
Control
You see what is kept about you, in the app, without having to ask.
Deletion
Erasing is a gesture, not a procedure. The account and the memory leave together.
Independence
If a provider shuts its door, the knowledge base stays open. That is what building on public data rather than a private database changes: nobody can pull the ground out from under us.
Do you run a museum, a monument, a festival? The editorial framework, control over the stories and the integration are decided with you.
See the venue offerThe first cities are coming.
Join the list for early access. No spam — just one message, when it's ready.