The short answer
Six Roosta missions use the camera. Every one analyses the live frame on your device and discards it immediately. No image is written to storage and none is transmitted. That is an architectural constraint in the app, not only a statement in a privacy policy.
What actually happens during a camera mission
The camera preview runs, an on-device vision model looks at each frame for the target — a mug, a dog, open sky — and returns a label with a confidence score. When confidence passes the threshold, the mission is marked complete and the frame is gone. Nothing is saved between frames and nothing accumulates.
- No frame is written to disk, the photo library, or a cache.
- No frame is transmitted anywhere. The missions work with the network off.
- No reference photo is stored from setup — Roosta names an object rather than asking you to photograph a place in advance.
- The camera permission is requested only when you first enable a camera mission.
Why this differs from photo-matching alarms
A common design in this category asks you to photograph a specific location during setup — your bathroom sink, your kettle — and then to reproduce that photo to dismiss the alarm. That requires keeping your reference photo, which means an app holds an image of the inside of your home indefinitely.
Roosta names a category of object instead. There is nothing to store, and the target can change every morning, which also makes it harder to pre-stage the answer the night before.