Let’s Talk
#33Consumer AI / Personal data20260 errors and 0 warnings in the vendor checker

Came True: Find Which Saved Wishes Came True in Your Own Photos and Documents

Built in-house

Retailbox R&D: built in-house for the UGREEN Hackathon 2026 on AI NAS, 2026

A skill for UGREEN's AI NAS that finds the wish screenshots saved on it and shows which of them already came true, with the ticket, certificate, order or photo where each one became real.

UGOS skill package (Markdown)ug-photoug-documentug-knowledgeug-filemgrLocal vision models

What was the problem?

A wish saved as a screenshot and the proof that it came true end up in different places on a NAS. A picture of a beach saved in winter, boarding passes in the documents by summer, photos from that shore in the gallery. The NAS's built-in search and UGREEN's 30 official skills don't link pictures to wishes; each uses one NAS expert.

How did Retailbox solve it?

The skill finds wish screenshots and plan notes, looks for later proof and shows 'wanted, came true' pairs to confirm. It works through four NAS experts: photos (search, metadata, vision), documents (question answering and extraction), files (tags) and knowledge (saving text). First comes a metadata filter: the owner's own camera, dated after the wish. Then the AI is asked twice, in different words, and a pair counts only if both answers say yes. Places are matched by photo location first, and then the AI checks the photo itself. A downloaded photo of the same beach doesn't count, and the owner has the final say. With permission it saves a 'Came True' book of the wishes that came true and tags the proof; the next run checks only what's new. It asks permission only to save the page, tags and album, and uses no outside services; per UGREEN's FAQ, the built-in Uliya models may use Uliya's cloud.

This is an early build, and the page says so. It has not yet been tested on a UGREEN NAS: the copy filter, the picture questions, finding screenshots, Russian text, capture dates and places are still untested there. The planned test is 7 of 7 pairs found and 0 of 3 fakes counted.

Facing the same problem? These are the services that cover this kind of work: AI agent development services.

What were the results?

  • ✓UGREEN kit 0.7.7 checker on the skill package: 0 errors, 0 warnings
  • ✓'Is this a wish?' on held-out screenshots, with three open vision models run locally: 14 of 14 for two models, 12 of 14 for the third
  • ✓A document as proof (delivered order, ticket, booking, certificate): 10 of 10 for all three models
  • ✓On the UGREEN NAS itself: not tested yet

Have a similar problem?

30-minute call. Free. We'll tell you honestly if we can help.