Privacy-first vocabulary capture

A tool that can capture a word from anything you're reading — even text you can't select — is genuinely useful, and that same power is exactly why its limits have to be clear. A private vocabulary app should be able to tell you precisely when it looks, what it keeps, and what it sends, and the honest answer for Capecho is: only when you ask, only the word and sentence you approve, and as little as possible beyond your own device.

Capecho recognizes the text using macOS's built-in, on-device text recognition — the same engine behind Live Text — so the recognition itself happens locally as a local OCR vocabulary app, not on a server. It runs only at the single instant you press the capture shortcut. There is no continuous, background, or scheduled reading of your screen, and the system returns only the recognized text — the screen image itself never reaches Capecho.

macOS gates this kind of capture behind a permission it labels "Screen Recording," and that name can sound alarming. It's just the operating system's mechanism for letting an app read on-screen text on demand; Capecho never records or streams the screen, and never watches it between captures. The permission is the door, and Capecho only opens it for the split second you ask it to.

Nothing enters your vocabulary library automatically. Every capture opens a preview you can edit, not just confirm: correct a mis-grabbed word, tighten the context sentence, or delete a stray email address, name, or account ID that happened to sit near the word. Only what you approve is saved.

This is the difference between a capture you can trust and one you can't. The word you grabbed is the unit you keep, and the sentence and its gloss are yours to adjust — so a private detail that wandered into frame never has to become a saved card.

The free word explanation is generated once from the word alone and shared through a public cache, so your sentence is never part of what builds it. Only when you ask for the meaning of a word as used in your specific sentence does that one sentence get sent to be explained — it's an explicit, per-word action you choose, capped at ten a day, never a background upload. If you'd rather keep more on your machine, you can simply not request that in-context layer; everything else still works.

If you would rather not grant screen-recognition permission at all, Capecho works in a copy-paste mode: it reads your copied selection only after you press the shortcut, and never monitors the clipboard in the background. This is a deliberate trust feature for a secure vocabulary app, offered as a first-class path, not a crippled fallback — you can build your entire vocabulary this way and never enable screen capture.

Capecho's capture and review both live in the macOS app right now; that's the only shipped surface, and it's where your library stays. A phone companion for reviewing words on the go is coming, and when it arrives the same privacy posture carries over — you, not a background process, decide what gets captured and explained. There are no hidden trackers and no data-collection business model underneath; the product is the vocabulary loop, not your screen.