The problem with most weather apps is that they tell you the temperature, not whether you specifically will be cold. A 14°C morning means something very different to someone who runs cold versus someone who's always warm. We wanted to close that gap.
Samukunai ('Are You Cold?' in Japanese) is the result. The premise: give it 10 feedback taps rating how you actually felt on a given day, and the morning push notification shifts from a raw temperature to something like 'on the cold side for you.' Same weather data — shaped around your personal cold tolerance.
The engine: percentile statistics, not ML
We didn't reach for machine learning. The personalization is pure 75th/25th percentile statistics: once you've logged enough feedback, the app knows your comfort thresholds. Today's temperature, humidity, and wind run through what we call the ComfortEngine, and the result tells you where today sits relative to your personal range.
This was a deliberate choice. An ML model trained on 10 data points is noise. Percentile stats work reliably with small samples, are fully explainable, and run entirely on-device. The stack: WeatherKit for live conditions, SwiftData + CloudKit for persistence, iOS 26 Liquid Glass for the UI.
Honest about the cold-start
Below 10 feedbacks, the app has nothing to offer that a standard weather app doesn't. There's no widget yet, and iOS 26 is a hard requirement. We shipped it anyway because the experience works once you're past the threshold.
If you're building a personalization feature, being explicit about 'you need X inputs before this gets useful' is better than letting users hit the cold-start wall without warning.
Who it's for
People who run cold. People who overdress. People who get sick every change of season and wish the forecast reflected that. Free, no ads, no in-app purchases.
App Store: https://apps.apple.com/jp/app/id6762537476
Top comments (0)