Auto-Tagging

Updated

How Tagging Works

When Recall indexes a screenshot it runs two classification passes:

  1. Apple Vision image classificationVNClassifyImageRequest assigns visual labels (e.g. “nature”, “document”, “map”) with a confidence threshold of 70%.
  2. OCR text heuristics — the extracted text is scanned for patterns that identify specific screenshot types.

The results are merged, deduplicated, and stored as a comma-separated tag list. Tags appear as sidebar filters and as chips in the detail panel.

Built-In Tags

TagHow it’s detected
codeOCR text contains func, class, struct, let, var, import, return, =>, def, or #include
receiptOCR text contains a dollar amount pattern (e.g. $49, $ 9.99)
webOCR text contains http:// or https://
errorOCR text contains error: or exception (case-insensitive)
mapOCR text contains latitude, longitude, directions, navigate, or street view

Additional tags may appear from Apple Vision’s visual classification — for example document, food, or nature.

Browsing by Tag

The AUTO TAGS section of the sidebar shows every tag present in your library with a count. Click any tag to filter the gallery to only screenshots with that tag.

To clear the filter, click All Screenshots at the top of the sidebar.

Source App Detection

In addition to content tags, Recall identifies which app produced each screenshot. It first reads the kMDItemCreator Spotlight attribute from the file’s metadata, then falls back to filename patterns for standard macOS screenshots (e.g. Screenshot 2026-04-30 at 10.23.45.png).

Source apps appear in the SOURCE APPS section of the sidebar. Clicking an app filters the gallery to only screenshots from that app.

Uncategorised Screenshots

Screenshots with no tags appear under Uncategorised in the sidebar. This is useful for reviewing images that haven’t been classified yet (for example, very new files whose Vision processing is still in progress) or images whose content Vision couldn’t identify.