Viaondo uses AI for the slow parts of building a trip: reading confirmations, drafting the days and checking places. It never decides what a traveler sees. Everything it adds lands in the agency’s draft, marked with where it came from, and travelers only see a trip once the agency publishes it.
Where AI is used
- Reading booking documents. PDFs, emails, screenshots and Word files become bookings, travelers and documents in the trip. Each booking keeps a link to the document it was read from, and what the AI couldn’t read for sure is listed under “To confirm”.
- Making a trip from a description. A few lines from the agency (“Lisbon and Porto, 8 days, a couple in May”) become a first draft of days and plans.
- Checking places. For each place a trip names, Viaondo looks for the place’s own web page and reads its address, local name, hours and phone. Facts it confirmed show the page they came from; the rest stays “to confirm”.
- Writing. A short programme for each day and the destination guides (getting around, food, good to know), when the agency asks for them.
- The trip assistant. In the dashboard, the agency can ask for changes in its own words (“move the museum to Thursday”). Each answer is applied as one change the agency can keep or undo.
The traveler app itself uses no AI. Phrases are spoken by the phone’s own voice, and the weather, maps and offline copy are not AI.
Who runs it
The models run on NaN (nan.builders), operated by Helmcode S.L. in Spain, on infrastructure in the European Union. They are open-weight models: GLM for reading documents and writing, Qwen for reading and for scanned pages, Gemma for short lookups and Whisper to turn a dictated description into text. NaN is listed among our sub-processors and covered by the data processing agreement.
To find a place’s web page, Viaondo uses TinyFish search. It receives only the place’s name, its city and, when a booking prints one, its address or supplier: never travelers’ names or the agency’s.
What the models receive, and what they keep
- They receive what the feature needs: the documents the agency uploads (with the names printed in them), the agency’s description of the trip, or the parts of the trip a question is about.
- NaN doesn’t store or log the documents or the answers, and doesn’t use them to train models. Neither does Viaondo.
- What was read stays in the agency’s own data, and the documents themselves are deleted 30 days after the trip ends.
- We never log document text or what a model answered.
Guardrails
- Nothing goes live on its own. AI writes into the draft; publishing is always the agency’s action.
- Every booking shows its source, so a fact can be checked against the voucher it came from.
- Instructions inside a document are ignored. A model is told to return data only, and its answer must match a strict format or it is thrown away.
- Times are checked by code, not by AI. Time zones, daylight-saving changes and meeting versus departure times go through the same checks as anything typed by hand.
- Limits. Each agency can import up to 300 documents a month with AI (fair use).
Your choices
An agency can build every trip by hand, without AI: the import, the assistant and the writing tools are only used when someone asks for them. Free place photos, which are not AI, can be turned off in the agency’s settings.
Questions about how AI handles your data: info@viaondo.com. The security page covers the rest of the service.