Inspiration
Many families keep cameras in an ageing parent's home, but camera apps alert on every motion. Families mute the alerts and still miss the quiet signals: Mum not reaching the kitchen by mid-morning, or a stranger returning to the gate night after night.
What it does
Ask "How was Mum today?" and Quiet Watch reads 14 days of camera events on the UGREEN AI NAS, learns each known person's routine per room, and compares today against it. The report lists normal activity, deviations and unfamiliar visitors, each with a snapshot so the family can check for themselves. It describes patterns; it never diagnoses or accuses.
How we built it
Quiet Watch is a Markdown-only Uliya Skill. SKILL.md orchestrates the security-monitor agent (cameras, event-timeline, identity-resolve, event-evidence) and the knowledge agent (save-text). The baseline is the share of the last 14 days each person was active in each room and hour; a slot active on 70% or more of days that is empty today becomes a flag.
Challenges we ran into
- This platform version has no scheduled triggers, so Quiet Watch is a check-in on request, not an alarm.
- Absence is harder to detect than presence: the Skill must tell "nothing happened" apart from "a camera was offline".
- Camera data is sensitive, so all processing stays on the NAS and the monitored person's consent is required.
What we learned
The value of home cameras is not more alerts, but fewer and better ones that understand a household's normal.
What's next
Validate event fields and query speed on a real iDX6011 Pro, then tune the thresholds against real routines.
Built With
- markdown
- ugos
- ugreen-ai-nas
- uliya-skill
