Launched this week
Redaktr
Catch what you're about to leak
10 followers
Catch what you're about to leak
10 followers
Redaktr is an on-device screenshot protection layer for Mac that catches secrets, customer data, faces, QR codes, and policy terms before screenshots are pasted into AI tools, Slack, email, support tickets, or docs.
How does it actually catch things on-device without sending the image to a server first? Curious whether the detection is fast enough to not mess with my normal screenshot flow.
@c_kurtar55949 Good question, this is the part I spent the most time on. It uses Apple's Vision framework for OCR and face/QR detection, which runs on the Neural Engine, plus a set of deterministic pattern matchers on top of the recognized text: known key formats (AWS, GitHub, Stripe, etc.), Luhn validation for card numbers, entropy checks for generic secrets. No model calls, no server, the image never leaves the machine. You can verify with Little Snitch if you want, the scanning path makes zero network requests.
Speed: a typical full-screen screenshot scans in under 2 seconds on Apple silicon, usually faster. Scan runs automatically on import so the flow is screenshot, glance at the boxes, export. If you use the folder watch mode it scans in the background and only interrupts you with a notification if it actually found something, so your normal capture flow doesn't change at all.
How does it handle screenshots that already exist in my clipboard before I install it, and does it work retroactively on images saved to disk?
@sattapanbmta Both work, slightly differently.
Clipboard: there's a Paste Image command (Cmd-Shift-V) that imports whatever image is currently on your clipboard and scans it, so anything sitting there before install is one keystroke away from a scan. The optional clipboard protection mode is forward-looking, it watches for new image copies after you enable it.
Existing files on disk: yes, fully retroactive. You can open any PNG/JPEG/TIFF/HEIC, drag a batch of files in, or point it at a whole folder and it scans everything, then lets you jump straight to just the images with findings. A few people have run it over their accumulated Screenshots folder as a cleanup pass, which honestly surfaces some uncomfortable stuff.
One thing it never does: it doesn't crawl your disk on its own. It only touches files and folders you explicitly open or choose to watch.
Does it run entirely locally, or does it need to call out to a server to do the redaction? Curious how it handles that on a locked-down corporate machine with no internet.
@nkurumak28108 Scanning and redaction are entirely local, no server involved, so the core product works with zero internet. Airplane mode, air-gapped, doesn't matter. That was a hard requirement for me, a redaction tool that uploads screenshots defeats its own purpose.
The honest caveats for a fully offline machine: license activation needs to reach the license server once (it's a Lemon Squeezy key check, no account), and after that there's an offline grace window rather than indefinite offline validation. The update check also uses the network but is skippable. No analytics or telemetry at all, so there's nothing else trying to phone home.
If you're on a machine that can literally never touch the internet, email me, that's a real use case I'd like to support properly rather than hand-wave.
Finally a tool that runs the privacy check locally instead of shipping screenshots off to some cloud service. Love that on-device angle, feels like the right default for anything handling PII and customer data.
@laydaheme Thanks, that means a lot. It honestly felt like the only defensible way to build it. A tool whose whole job is catching leaks shouldn't create a new place for them to leak to. The nice side effect of local-only is that I never have to write a privacy policy paragraph explaining what happens to your images on my servers, because there are no servers and I never see them.
If you end up trying it, I'd genuinely love to hear what it catches or misses on your real screenshots. That feedback loop is the whole product right now.