AI Offline Tour Guide

A tour guide that lives on your phone, not in the cloud — so it works anywhere in the world, even offline, without needing to download location-specific guides. Open it in Paris, on a hiking trail outside Tokyo, or in a small town with no cell signal. The screen fills with places worth stopping at. Tap any of them and your guide tells you the story — its history, why it matters, what to look for. Walk around and ask follow-up questions in plain language, by typing or by voice. No login. No account. No tracking. It works offline because a small AI model lives directly on your device. Downloaded once (about 580 MB on smaller phones, larger and multimodal on newer ones), it runs entirely on your hardware. Your location, your questions, your photos — none of it leaves your phone. The offline mode is the real product: airplane, basement, foreign country with no roaming, dead-spot trail. It still works. When you're online, the app grounds answers in Wikipedia so the facts stay accurate, and shows you the source. When you're offline, the on-device AI takes over and is honest about it: every answer is labeled "ON DEVICE" so you know to double-check anything critical. The two modes blend automatically based on your connection; you can also pin to one in Settings. FEATURES • Tap any place on the "Around you" list to hear a short, grounded narration of its history and character. • Ask anything by typing or by voice, in your own language — the interface, voice, and AI answers cover 43 languages. • Auto-Guide: hands-free narration as you walk. Switch it on and it narrates notable places as you approach them (opt-in; uses background location, and stops the moment you turn it off). • Camera identify (on newer phones that run the multimodal model): point your camera at a landmark, statue, plant, or dish for an "Identified: …" answer and a short explanation. • "Plan my day" turns the next few hours into a walking itinerary built around the places that match your interests. • A map view lets you browse what's nearby visually, with the same tap-to-narrate flow. • The whole experience runs with no signal — flights, trails, subways, roaming-free travel. • Settings let you choose a topic focus (history / nature / food / culture / hidden gems), a narration length (short / standard / deep dive), a voice for spoken responses, distance units, and the search radius. PRIVACY • No account, no login, no analytics SDK. • Location, voice, photos: processed entirely on your device. • The on-device AI model is downloaded once and runs locally thereafter. • When online, the app contacts Wikipedia (facts), OpenRouteService (walking distances), Google Maps (the map view), and Hugging Face (only for the one-time model download). When offline, it contacts nothing. • No personalized advertising. No third-party tracking. • Optional anonymous crash diagnostics (Sentry): stack traces only — no user content, no identifiers. ABOUT THE AI • On-device LLM: Google's Gemma family (3 1B IT int4 on smaller phones, 4 E2B IT int4 multimodal on newer ones; a larger 4 E4B model can be chosen on high-RAM devices). • Inference runs on-device via Google's LiteRT runtime — no server, no round-trip. • Answers are grounded in your live location. Online, the app draws facts from Wikipedia; fully offline, it uses the bundled place database (GeoNames, ~150k places) plus the model's own pretrained knowledge — labeled "ON DEVICE," with a reminder to double-check important facts. • The model's knowledge is not guaranteed to be accurate, especially for lesser-known places — treat its answers as a starting point, not a definitive reference. REQUIREMENTS • Android 7.0 or newer (API 24+). • Wi-Fi recommended for the first model download (~580 MB to ~3.5 GB depending on device). • 4 GB RAM recommended; the app picks a smaller, faster model on lower-RAM devices and a larger multimodal one on higher-RAM devices. If you're walking somewhere with a story you don't know, this app is for you.