Every time I wanted to show someone a video of my dog eating carrots from a bowl (one piece at a time, crunching for ten seconds each, it's genuinely the funniest thing), I'd end up scrolling through thousands of photos trying to find it. Apple Photos couldn't help. I tried searching "dog eating carrot" and got nothing. That frustration turned into a question: why can't I search my photos the way I actually remember them? Not by date or album, but by describing what happened. "Food in Berlin last month." "Cherry blossom." "That screenshot of the flight confirmation." We started building Memorable to solve exactly this. The biggest constraint we set for ourselves was that everything had to run on-device. Your photo library is the most personal thing on your phone. Sending it to a server for AI processing is a trade most people shouldn't have to make. That constraint shaped every technical decision. We chose SigLIP2 (Google's vision model), converted it to Core ML, and got it running on the iPhone's Neural Engine. We built a query parser that breaks a sentence like "food in Berlin last month" into three separate filters: what it looks like (food), where it was taken (Berlin), and when (last month). We used Metal Performance Shaders for GPU-accelerated search across thousands of embeddings in milliseconds. The hardest parts were the ones we didn't expect. Location names stored in German on a German-locale phone wouldn't match English search queries. The vision model produces much lower similarity scores than we were used to, so every threshold had to be recalibrated from scratch. We're launching today with photo, video, and screenshot search, plus features like Find Similar and duplicate grouping. Free to try with 500 photos, Pro for $4.99 one-time. We'd love your feedback. Especially queries that should work but don't. We're still making it better.

Memorable

Photo search that works like your memory does.

Photo search that works like your memory does.