Brent Haskins / Applied AI
Screenshot search fails without Finder tags and date groups
OCR text search alone drowns users in matches. Brent Haskins built Shelf Studio to combine Apple Vision indexing with Finder tags, date grouping, and smart folders—so screenshot libraries feel browsable, not like grep through noise.
Search that returns four hundred matches is not search—it is guilt.
Shelf Studio indexes screenshot text on device, but the daily win is narrower: “show me red-tag captures from last sprint with checkout in the image.”
Metadata users already maintain
Power users tag in Finder. Casual users at least remember “last week.” Ignoring that and relying only on OCR recreates Desktop chaos in a prettier grid.
Date groups as first-class filters
Screenshots cluster by capture time. Rolling weeks beat infinite scroll for performance and cognition. Auto-refresh when the watch folder changes so new shots appear without reload buttons.
Batch actions earn retention
Rename, tag, delete in bulk—operations Finder does awkwardly on hundreds of PNGs. Inline preview prevents mistaken deletes.
Pair with OCR honestly
Vision errors on UI chrome and small type. Show confidence implicitly by ranking and recency; let users edit mistaken tags.
Related work
/projects/shelf — native Mac app. Companion: /blog/on-device-ocr-desktop-product-engineering.
Brent Haskins ships desktop utilities where the file system is the database. If you are designing capture tools, design for filters users already understand.
FAQ
Questions people ask about this topic.
Why combine OCR and Finder tags for screenshot apps?
OCR finds text inside images but returns too many hits for common words like error or home. Tags and date groups mirror how people already organize Mac files. Filters stack: tag plus date range plus text token narrows results faster than search alone. Users trust results they can explain.
What makes Shelf different from cloud screenshot tools?
Shelf keeps files on disk, indexes locally, and refreshes when folders change—no upload queue. It targets designers and engineers who capture UI bugs and references hourly. Privacy and speed are the positioning; feature depth is grid browse, batch rename, tag, delete, and inline preview.
Sources