Screenshot OCR on Mac, without a round trip to a server.

The capture

• Region or window – drag a selection, or press Space to click a whole window instead. ⏎ takes the entire screen.
• Image or text – T switches what you get. In text mode the recognized text is copied to the clipboard and saved as a text element, and no image file is written to disk at all.
• The same modifiers as the system tool – hold ⇧ to lock one side mid-drag, ⌥ to resize from the center, Space to move the whole selection.
• A loupe that reads pixels – it follows the cursor at 4×, 8× or 16×, with the pixel you are on outlined in the middle, so an edge can be lined up exactly.
• A color under the cursor – Tab copies it and keeps it in the stream as text, so you can find that color later by searching its own digits.

What the OCR buys you

Recognition runs in the background right after the capture and attaches the text to the image, so a screenshot of an error, an invoice or a slide is searchable by what it says – not by a filename you would have had to invent. The same text feeds search by meaning, which means a screenshot can answer a query that shares no word with it.

Nothing about this needs a network. Vision is part of macOS and runs on the machine; the embedding model behind semantic search runs on your Mac's own GPU.

After the capture

Opening a capture turns the window into a small image tool: crop with pixel-exact handles, resize to an exact width and height, rotate in 90° steps, zoom to 6400% where individual pixels are drawn as squares, and pick a color with Tab in HEX, RGB or HSL. A backdrop – padding, background, rounded corners, shadow – can be baked under the image for the screenshots you are about to share.

Permissions, once

Gisti asks for Screen Recording the first time you take a screenshot – macOS requires it for any app that reads the screen. Nothing else is needed: the global hotkeys work out of the box, and no capture ever leaves the Mac.