Inspiration
Every household has a pile of bills, receipts, warranties and insurance policies. Many of us scan or download them to our NAS, but then they sit scattered across folders. Payment dates get missed, policies lapse, and when an appliance breaks nobody can find the warranty, or work out whether it's still covered.
What it does
Home Paperwork Keeper is an AI Skill for UGREEN AI NAS. It works on documents already stored on your NAS, so there's nothing to upload by hand. Ask in plain language, for example "What's due this month?" or "Is my laptop still covered?"
1. Household document intelligence. It links related documents and reasons across them. A laptop receipt dated 2025-01-10, a 2-year manufacturer warranty and a 1-year extended warranty become one item, "Laptop: covered until 2028-01-10", with the calculation and all three sources shown. Ambiguous terms, such as "from date of delivery" with no delivery date, are flagged rather than resolved silently.
2. Verify, don't guess. Every fact carries its source document. When two documents disagree, for example two different renewal dates for one insurance policy, the Skill shows both values with their sources and asks you to confirm. It never picks one silently, and it never invents a missing date.
3. Deadline digest. Items that need your check come first, then overdue, due soon and expiring items, each linked to the original file. The digest is saved as a note on the NAS.
4. Confirmation-gated tidy-up. Optional tags and a clean folder layout, applied only after you approve. It never deletes or overwrites anything.
How we built it
The Skill follows the Uliya Skill Development Standard. SKILL.md defines a staged workflow orchestrated through registered UGOS expert agents, using only supported CLI capabilities:
- Find: Files expert agent (find, browse, deduplicate)
- Read and extract, with sources: Documents expert agent
- Link and reconcile: Documents expert agent (coverage reasoning, conflict detection)
- Organize (optional, confirmation-gated): Files expert agent (tag, move)
- Report: Knowledge expert agent (save text)
Supporting references define coverage-reasoning rules with a worked example, uncertainty handling (confidence levels and conflict cards), the extraction schema, digest format, folder layout and the finalist test plan.
Where AI processing runs, on the device or through a cloud AI service supported by Uliya, follows the user's own Uliya model configuration. The Skill adds no external services of its own.
Challenges
- Designing coverage rules that are useful but honest about ambiguity.
- Making uncertainty a feature: a wrong date is worse than no date.
- Keeping every file change behind an explicit confirmation step, and masking account and policy numbers to the last 4 characters.
What we learned
Trustworthy AI on a NAS is less about the model and more about workflow design: reasoning across documents, showing evidence, and keeping the user in control.
What's next
As a finalist, we would test on the UGREEN AI NAS iDX6011 Pro with about 30 synthetic documents: a deadline-digest test, a linked-coverage test (receipt + warranty + extended warranty), and an edge case with conflicting, corrupt and illegible documents.
Built With
- ai
- markdown
- nas
- ugos
- ugreen
- uliya
