Inspiration
Every screenshot holds something worth keeping: a booking code, an address, a recipe. Then it sinks among memes. Screenshots are also deeply private (OTPs, bank apps, IDs), so sending them to a cloud AI is a non-starter.
What it does
Snapsift reads the screenshots your phones back up to the NAS and sorts them into 11 buckets: bookings, addresses, recipes, receipts and more. It makes an album for each bucket and builds a searchable index. Passwords and IDs are tagged and never repeated. Ask "find the screenshot with the clinic address" and it answers with the file and the matching line. Nothing is deleted, and every change needs your approval.
How we built it
A Uliya Skill that hands the work to UGREEN's on-device agents: photo-agent reads and sorts, file-agent tags and finds duplicates, knowledge-agent saves the index, and media-agent cit test-first:
Challenges we ran into
An independent review found two critical bugs. Re-runs never got past the first batch, and search could quote an OTP in chat. Both are fixed. The platform has no locked albums and n around both.
Accomplishments that we're proud of
Privacy-first: no screenshot leaves the NAS, secret is deleted, and every change needs approval.
What we learned
What a screenshot says matters more than how much a viewer catches what your own tests miss.
What's next for Snapsift
Testing on a real UGREEN NAS, OCR for Chinese and albums if the platform adds them.
Built With
- claude
- node.js
- ugreen-nas
- uliya-skill-kit
