Inspiration
Freelancers and tiny agencies receive client files all day on WhatsApp, Telegram and email. The files pile up unsorted on phones and laptops, invoices get lost, duplicates get paid twice, and nobody can see what arrived today.
What it does
Intake Desk triages a freelancer's NAS inbox. A single spoken command — 'Triage my intake inbox.' — triages the whole inbox on the NAS itself. It classifies new client files, extracts key facts, transcribes voice notes, files everything into dated client folders, and writes a plain-language daily brief — after the user approves the plan.
The Skill lists new files, finds duplicates, extracts facts from documents, transcribes voice notes, builds a filing plan, shows the plan and draft brief for approval, then files everything and saves the brief and ledger. Nothing moves or deletes without approval, and the Skill never deletes; the user clears Duplicates by hand.
Receipt filers sort one file type by month. Intake Desk handles mixed intake — documents, receipt photos, voice notes — names files by client, flags duplicate payments and missing invoice numbers, and produces a decision-ready daily brief instead of just a tidy folder.
How we built it
Intake Desk is a Uliya orchestration (SKILL.md plus references) delegating to four registered expert agents: file-agent (ug-filesystem, ug-filemgr, file-workflow), document-agent (ug-document), knowledge-agent (ug-knowledge) and photo-agent (ug-photo). Each step passes only the real result of the previous step onward. All AI runs on the NAS local model; no file leaves the device; moves need platform approval.
UGREEN AI NAS features used, by step: file search and duplicate scan list the inbox; document extraction and QA read invoices and contracts; photo OCR reads photographed receipts; knowledge transcription turns voice notes into text; the file workflow planner builds the filing plan; tags and the knowledge saver write the brief and ledger.
Built: the full SKILL.md workflow, classification and naming rules, brief template, approval flow, and a local Python prototype of the rules with 13 passing unit tests on sample files. Not built: nothing in the Round 1 package. Needs NAS testing: real extraction accuracy, OCR on phone photos, transcription quality, and the approval flow on device.
How we will test it as a Finalist
Core 1: 20 mixed files (invoices, receipts, contract, photos, voice notes) — check every file lands in the right client folder with the right name. Core 2: duplicate invoice — check it is flagged, not paid twice, and moved to Duplicates. Edge: a blurry photographed receipt with no readable total — check it goes to Needs Review, never guessed.
Challenges and limits
Extraction depends on the NAS local model, so handwriting and blurry photos may fail and land in Needs Review. Non-English voice notes depend on transcription language support. No outside service is used, so there is no cloud dependency. Third-party components at runtime: none. Authoring tool: UGREEN Uliya Skill Kit v0.7.7. Original work by Asadullah, 2026.
Built With
- ai
- local
- nas
- python
- ugos
- ugreen
- uliya
