I've been building Fog, a notes app that tries to solve a problem I kept running into: notes apps either make you organize everything manually (folders, tags) or don't organize at all.
Fog uses Apple's on-device Foundation Models framework (the same one powering Apple Intelligence) to read your notes and group related ones into "Clouds" automatically no manual tagging, and nothing ever leaves your device since it's all on-device inference.
The interesting technical challenge was the clustering logic. I ended up using a union-find (disjoint set) structure combined with NaturalLanguage embeddings to figure out which notes are actually related, then merge them into clusters that update as you add more notes. Getting the similarity threshold right so Clouds don't over-merge unrelated notes or fragment obviously-related ones took a lot of tuning. Demo video (56 sec, shows the auto grouping) Curious if anyone else here has worked with Apple's on-device Foundation Models yet how are you handling similarity/clustering tasks, and did you run into similar tuning headaches with thresholds?
I struggle with organizing my notes, creating folders, figuring out what goes where, selecting things manually… the list goes on. So I built a simple app that aims to solve this with 4 features: Auto-naming notes, Auto grouping into clouds, Auto cloud grouping, and Ask anything about your notes
Powered by Apple’s on-device Foundations Model (requires a device with Apple Intelligence). Your notes sync across your devices via iCloud, so they stay in your Apple ecosystem.
