Text that was never text
A modal, a scan, a chart, a slide in someone else's screen share. If it's on your screen you can draw a box around it, and what comes back is characters you can paste.
Capture
Drag a box around it,then just ask
A slide in someone else's screen share, a scan, a chart, an error nobody can copy. Khint reads the region you drew and hands you the text, the answer, or the ticket.
One box,
two rows in the palette
You decide before you draw. Extract text opens nothing and puts plain characters on your clipboard. Capture & ask pins the image to a conversation, and that conversation can end in your tools.
KhintPalette
Sprint 24
/
Capture
The text nobody can select
A scan has no text layer, a screen share has no file, a chart has no copy button. Khint reads the pixels and hands you characters. The overlay tells you how many landed, and nothing else opens.
Made for the things
you can't select
Copy-paste stops at the edge of a text field. A capture doesn't care whether the words are text, pixels, someone else's window or a file you were sent.
Or an image you were sent
PNG, JPG, WebP, HEIC, GIF, BMP, TIFF. Drop it on the Capture page and it takes the same path as a region you drew.
Ask again without capturing again
The image stays pinned for the whole conversation, so a follow-up is a follow-up and not a second capture.
It can stay with the rest of your work
One button writes the whole thread into your active Memory session, so the reasoning sits beside the ticket it produced and the agents you run later can draw on it. Ask three more questions, press it again, and only the new exchanges are added.
It sees the box you drew
and nothing around it
A capture is a gesture, not a permission. Nothing is watching between two of them, and there's no recording to search later.
Security and privacyOnly the box you drew leaves the Mac
No background recording, no screen history
Screen Recording permission is never asked
The macOS selector draws the region
One capture at a time, by design
Cancel a run and nothing is written
A saved thread lives in local SQLite
Nothing trained on your captures
Three ways to start a capture
Press the shortcut, hand it a file, or let a workflow draw the box for you.
From the palette
⌘⇧K, then the row you want. Capture has no global shortcut of its own, and doesn't need one.
- Works over any app, full screen or not
- About one credit a capture
- Escape before the box is drawn costs nothing
- macOS 13+ or Windows 10 22H2+
From a file
The screenshot someone sent you, a photo of a whiteboard, a page you scanned. Same path, same result.
See document agents- Seven image formats
- Drop it, or pick it from disk
- Recent captures stay on the page
- A PDF with a text layer never needs it
As a workflow's first step
Set a workflow's input to Capture and one shortcut draws the box, then files everything that follows.
See workflows- Up to five steps behind one shortcut
- Starts from · Capture
- A step can fan out to three tools at once
- Every write lands back in your session
The rest of Khint
A capture is an input. These three are what happens to it next.
Common questions about screenshot to text on Mac
How do I copy text from a screenshot on Mac?
Open the Khint palette with ⌘⇧K and pick Extract text. Drag a box around the region and the plain text lands on your clipboard, ready to paste anywhere. No window opens and nothing else changes. Image files work the same way: drop a PNG, JPG, WebP, HEIC, GIF, BMP or TIFF on the Capture page.
Can Khint answer questions about what's on my screen?
Yes. Pick Capture & ask instead, and the region you drew is pinned in a chat window. Ask about it, then keep asking: the image stays in context for the whole conversation, so a follow-up never means capturing again. It reads charts, scans, slides in a screen share and modal dialogs, none of which you can select.
Can Khint act on what it sees, not just describe it?
A "Do this" button sits under the last reply and at turn zero. It resolves a plan out of your own agents and connected tools: draft the bug report, then create the Jira issue, post to Slack. You see the steps before anything runs, the pickers open before the model call so canceling costs nothing, and the link to what it created lands on your clipboard. If the run was good, Khint offers to keep it as a workflow.
Does Khint record my screen in the background?
No. A capture is a gesture: nothing happens between two of them, there's no recording to search later, and the macOS region selector draws the box, which is why Khint never asks for Screen Recording permission. Only the region you drew is sent, and it's not kept on our servers.
Do my Capture & ask conversations disappear when I close the window?
Not if you keep them. Save to session writes the whole question-and-answer thread into your active Memory session, so the reasoning stays beside the ticket it produced and later agents can draw on it. Ask a few more questions and press it again: only the new exchanges are added, never duplicates. Sessions live in a local SQLite file on your Mac and never sync.
Is there a free menu-bar OCR app for Mac?
Yes. Khint's free tier includes 300 credits a month and a capture costs about one credit, drawn from the same wallet as your AI actions. Paid plans give you a bigger one: 900 credits on Lite, 4,000 on Pro, 7,000 on Max. It lives in your menu bar behind ⌘⇧K and runs on macOS 13+ or Windows 10 22H2+.
Try Khint today
Free with 300 credits a month. No credit card. macOS 13+ or Windows 10 22H2+.