Inspiration
At home we scan every letter that matters: bills, tax notices, health statements, insurance renewals. The scans end up in one folder with names like scan_0042.pdf. Finding last year's tax notice takes ages, a payment deadline hidden in a letter is easy to miss, price rises go unnoticed, and nobody knows which old papers can go.
What it does
Paper Trail is a Skill for UGREEN AI NAS with two modes.
Sort mode. You say "sort my scans". It reads every new PDF and every phone photo of a letter in your inbox folder, works out what each one is (bill, tax, health, bank, insurance, housing, work, other) and shows you a plan: where each file will go, its tags (to-pay, to-answer, keep, check-by-hand), the amount and due date it found, how a bill compares with the previous one from the same sender, and how long to keep the paper. When you say yes, it files everything and saves a monthly digest: what to pay or answer and by when, which bills went up, and which old papers are past their keep-until date.
Ask mode. Later you just ask: "how much did I pay for power this year?" or "where is my 2025 tax notice?". It answers from your own papers and gives the files it used.
Nothing is deleted, nothing is overwritten, and your tax and health papers never leave the NAS. Old papers are only listed; you decide.
How we built it
With the official Uliya Skill Kit. The Skill is a SKILL.md that delegates each step to the NAS experts: file-agent to list, tag and move, photo-agent for OCR on photos, document-agent to extract sender, date, amount and deadline, find the previous bill and answer questions, knowledge-agent to save the digest. Each step only uses the real result of the step before. Keep-until dates come from a small rules table in references/.
Challenges
Writing a workflow that never guesses. If a photo is too blurry or a letter has no date, the file is flagged check-by-hand instead of getting a made-up amount.
What's next
If selected as a finalist: run the test plan on a real NAS with the set of fake bills and letters we already made, check OCR on phone photos and the extract output on French and English documents, then record the demo.
Built With
- markdown
- ugos
- uliya-skill-kit