How it works
Three steps, all on your device:
- OCR: the app reads the text in your screenshot with Tesseract, an open-source text-recognition engine that ships with the app. The image is upscaled 2x first for accuracy, and every recognised word keeps its box on the image.
- Pattern checks: the recognised text is checked against patterns for emails, phone numbers, Social Security numbers, card numbers (validated with the Luhn checksum, so random 16-digit numbers are not flagged), AWS keys, GitHub tokens, Slack tokens, Stripe keys, login tokens, URLs with passwords inside, IP addresses, labelled secrets such as api_key = ..., and long high-entropy strings that look generated.
- Face detection: a small on-device face detector finds faces. It only draws boxes; it never identifies anyone, and it does not recognise faces.
Then comes the step that matters most: review. Every finding is listed with a checkbox (checked by default) and a plain-English explanation. Download and Copy stay disabled until you have seen the review list and ticked "I have reviewed every finding". The scrubbed PNG is built from the image on the page with solid black boxes or blur burned in. The original image is never modified.
What it catches
- Emails, phone numbers (US and international shapes), and IP addresses.
- US Social Security numbers in XXX-XX-XXXX form, with impossible area numbers rejected.
- Card numbers that pass the Luhn checksum (13 to 19 digits).
- AWS access keys, GitHub tokens, Slack tokens, Stripe live keys, and JSON Web Tokens.
- URLs with a username and password inside them.
- Secrets written next to their label, and long high-entropy strings that look generated.
- Faces, as boxes, so you can decide whether each one should stay.
What it can MISS
Reduces risk, never guarantees: always review before sharing. The scan can miss:
- Handwriting, tiny text, stylised or distorted fonts, and text at an angle.
- Faces at angles, partly hidden faces, and small faces.
- Secrets split across lines or columns, or partly covered by another window.
- Secrets in an image the OCR misreads by even one character (a 0 read as an O breaks a key pattern).
- Anything in a language other than English: the OCR model is English-only.
- Information that is not text or a face: a visible address on a letter, a reflection, a name on a badge in handwriting.
Your review is the real protection. The scan is the assistant that points at the likely problems. If you spot something the scan missed, use Draw a box to redact and add it by hand.
Mandatory review before copy or download
Most redaction tools let you blur first and look later. Screenshot Scrubber does it the other way round: the Download and Copy buttons stay disabled until you have viewed the review screen and ticked "I have reviewed every finding". That gate exists because the scan is fallible. A tool that redacts silently teaches you to trust it; a tool that makes you look teaches you to check.
If a secret already leaked
Scrubbing the screenshot is not enough once someone may have seen the secret. Rotate it: revoke the API key or token and create a fresh one, change the password, ask the bank to replace the card. For an AWS key, check the AWS console for unexpected usage. For a GitHub token, check for new keys or webhooks you did not create. Assume a shared secret is copied.
Built-in tests: a published catch rate
Press Run built-in tests in the app. It generates synthetic screenshots with planted secrets (email, phone, card, AWS key, GitHub token, Stripe key, Slack token, SSN, password URL, IP address), plus a clean image and a lookalike image (short key fragments, a bad-Luhn card, a 7-digit number), runs the full pipeline, and publishes the catch rate in the app. The test suite asserts at least 90 percent on planted secrets, zero false flags on the clean image, and zero on the lookalikes. Points and badges reward careful habits: First Scan, First Clean Share, Careful Reviewer, and a 7-Day Streak.
Your screenshot never leaves your device
OCR, pattern checks, and face detection all run in your browser. The page security policy blocks connections to every outside server. There is no account, no analytics, no recording, and no cookie. The built-in Offline proof panel shows a live log of every network request the page makes, lets you simulate a fully offline session, shows exactly what is stored on your device, and deletes it all with one button. The app works offline after your first visit. Your screenshots are never stored: closing the tab drops them.
More private, on-device tools: Little Hands teaches baby sign language, Fingerspell teaches the ASL alphabet, and Tone Trainer coaches Mandarin tones, all with the same nothing-leaves-your-device design.
This is a home practice tool, not a teacher, therapist or safety device. Everything runs on your device: no video, audio or images are sent anywhere. Feedback is automatic and can be wrong. For real instruction, learn from privacy and security experts, linked on the Resources page.