Inspiration
I keep years of family and personal photos on my phone, not costly iCloud; a NAS gives them a home. Phone search is nearly useless, so I always scroll to a rough date. In iPhone 18 Pro Photos (iOS 27.0.1), Korean searches for "2025 US barbecue" found nothing; "grill" found 1,289 mostly unrelated photos. My photo has no GPS.
What it does
TracePic finds an old photo from a vague memory, like "the barbecue on my 2025 US business trip", in my own NAS photos, plus family shared albums, and shows candidates with evidence. Photo search now works even without location data.
The barbecue photo cannot show it was the US, but other photos can, like flipping through my album thinking "it was around here". Without GPS, TracePic uses the capture time zone (GMT-04:00) and nearby photos from the same phone: my barbecue shows only meat, but a same-day grocery photo shows a US-only store brand and ounce labels.
- "Find the barbecue photo from my 2025 US business trip."
- TracePic confirms the clues.
- It offers the trip: "Apr–Jun 2025, GMT-04:00, US from nearby photos?"
- The user confirms.
- Up to 12 candidates appear with evidence.
- Not there? The user adds a detail.
How I built it
Pure-Markdown Uliya Skill: two modes, no scripts. The parent delegates self-contained tasks to photo-agent, file-agent and knowledge-agent, passing only real results. Photos never change; the only write is the private trip index, after approval. Nearby evidence stays within one phone. All five delegation states are handled.
Challenges
Time zones and nearby photos are evidence, not proof: spring GMT-04:00 also covers parts of Canada, the Caribbean and South America; neighbors can be hours apart. Screenshots lack offsets. The local vision model may misread small labels and is slow, so batches are limited. Same-model family phones can look like one. The index, a location history, stays private.
What's next
Pending NAS validation: offsets from meta, local capture times, vision throughput.
AI tools
Claude Code (Anthropic) was used for kit analysis, analyzing my phone search test screenshots, drafting and rule-checking the Skill Markdown, and drafting the proposal answers in English, then revising them from my own Korean answers about my experience and motivation; I reviewed the final text. The core idea, the nearby-photo inference, the index, the name and the motivation are my own.
Built With
- claude-code
- file-workflow
- markdown
- ug-filesystem
- ug-knowledge
- ug-photo
- ugos
- uliya
- uliya-skill-kit
