ShadowClip · Quiet Signals Lab

What it does

You bring ShadowClip an image — by dropping it into the app, sharing it from Finder or Photos, or using the Finder Quick Action on macOS. ShadowClip runs detection and shows you a review screen: a list of everything it found, overlaid on the image. You decide what gets redacted. When you save, a new image is written with opaque black fills over the regions you approved. The original is never touched.

Detection runs entirely on-device using Apple Vision for OCR, face detection, and barcode detection; the Natural Language framework for named entities; and hand-authored pattern matchers for credentials and structured data. The app has no network entitlement and cannot make outbound connections.

Developer Secrets Regex patterns derived from each issuer's published key format catch AWS access keys, OpenAI keys, GitHub tokens, Stripe keys, Anthropic keys, Slack tokens, JWTs, and 15+ others. Each pattern is matched against the raw and normalised OCR output, with confusable-character variants to handle common Vision OCR misreads (I/1/l, g/9, s/5).

Structured PII Credit card numbers (Luhn-validated), IBANs (ISO 13616 mod-97 validated), email addresses, phone numbers, IPv4 and IPv6 addresses, MAC addresses, social security numbers, national insurance numbers, and postal addresses via Apple's NSDataDetector.

Faces, QR Codes & Barcodes Apple Vision locates faces and barcodes directly in the image. These are auto-accepted — no OCR step required.

Named Entities The Natural Language framework identifies names, organisations, and locations in OCR'd text. These are surfaced as suggestions for your review rather than auto-redacted, because precision is lower than for structured patterns.

Click-to-Redact After detection, every word in the image is individually clickable. If the automatic pass missed something, click the word to redact it. For regions with no text — a handwritten note, a face in a logo — draw a rectangle anywhere on the image.

Clipboard & Save Copy the redacted image to the clipboard in one keystroke, or save it as a new PNG. An optional audit sidecar records what categories were redacted and a SHA-256 fingerprint of the output file.

Privacy by architecture

Your screenshots never leave your device. This is enforced at the architecture level: the app has no network entitlement in its sandbox, so the operating system will reject any outbound connection attempt regardless of what the code tries to do.

No account is required. No telemetry is collected. No crash reports are sent automatically — if you want to report a bug, you do it by email.

The input image is processed in memory and never written to disk. The only files written are the redacted output you choose to save, and the optional audit sidecar alongside it.

System requirements

macOS macOS 14 Sonoma or later. Works on Apple Silicon and Intel. Images come in via drag-and-drop, the Share Extension (Screenshot.app, Finder, Safari), or the Finder Quick Action.

iOS iOS 17 or later. Images come in via the Share Extension from Photos, Files, Safari, or any app with a share sheet.

Frequently asked questions

Does ShadowClip send my screenshots anywhere?

No. The app has no network entitlement and cannot make outbound connections. All processing happens locally.

Do I need an account or subscription?

No account. No subscription. One-time purchase on the App Store, covering both macOS and iOS.

What languages does text detection support?

The OCR and entity detection use Apple's Vision and Natural Language frameworks. Coverage is best for Latin-script languages. Credential pattern matching is language-agnostic — an AWS key looks the same in any screenshot.

Can I undo a redaction?

Yes, until you save or copy. The review screen has full undo. The original image is never modified; the redacted version is always a new file.

Does automatic detection catch everything?

No. The review step exists because no pattern matcher has perfect recall. Named entities in particular have meaningful false-negative rates on unusual names or non-Latin text. Always review the output before sharing.

Is Family Sharing supported?

Yes — one purchase covers all members of your Family Sharing group on both macOS and iOS.

Support

ShadowClip is built and maintained by Quiet Signals Lab. For bug reports, feature requests, or anything else, reach out directly.

Include your macOS or iOS version and a description of what happened. Screenshots are welcome — run them through ShadowClip first.

hello@quietsignalslab.com →