Skip to content

How narration and the guide conversation work

Every place in the catalogue carries several content fields (history, architecture, culture, food and so on). When the tourist reaches a stop, the app asks the server for a narration for that place and that profile: the server picks the lead field from the user’s interests (the first one with content), adds the rest and hands it to the language model with a system description of the guide’s role. The model only ever sees catalogue content — nothing sent from the app enters the prompt.

The default narration has six beats (hook, story, teach, sense, wow, bridge) and fits a few minutes of reading. The peak stop gets its own instruction: it is meant to be the strongest story of the whole tour.

The app is audio-first: the finished text is read by the device’s speech synthesiser while the text stays visible in the narration pane. Pause, the stop list and the transcript are one tap away for the whole tour.

When there is no internet or the server fails

Section titled “When there is no internet or the server fails”
  • If the tour was downloaded earlier, narration is read from the offline bundle.
  • If the server does not answer and there is no bundle, the app says a short fallback sentence with the place’s name — it never goes silent and never invents a description.

During the tour the user can ask about the place. The answer is built from the same catalogue content. Daily limits: guest 5 questions, free account 20, Premium unlimited. On Premium the conversation runs by voice through an agent that knows the current stop; outside Premium there is a text chat. A failed answer does not consume the quota.