Does my document get uploaded anywhere?

No. There is no server to upload it to. Everything — opening the file, reading it, covering things up and writing the new file — happens inside this browser tab, on your own machine. You can watch this for yourself: open your browser's network panel, redact a document, and you will see nothing leave.

You can also just pull out the network cable. The tool carries on working.

You do not have to take that on trust. This page carries a content security policy — a set of rules the browser itself enforces, not something the code can decide to ignore. View the source of this page and you will find it in the first few lines. The line that matters is connect-src 'self': the browser will block any attempt to send data anywhere but back to where this page came from, and nothing here asks it to. form-action 'none' means no form can post anywhere at all. If the code were ever changed to try, the browser would refuse the request and log the refusal in the console.

Is a black rectangle really enough?

On its own, no — and that is the reason this tool exists. Drawing a black box over a word in most editors leaves the word underneath it. Anyone can select it, copy it, or pull it back out of the file.

Blinded does not draw boxes over your document. It turns every page into a picture, paints the boxes onto the picture, and builds a new file from those pixels. The original text is not hidden in the result — it is not in the result at all. Nothing carried over from the original file is left to dig through — no original text layer, no fonts, no annotations.

Two things can put text into the finished file, and neither can bring back what a bar covers. Placeholder labels are written from the list of labels you can see. And if you tick Searchable text when saving, the pages are read after the bars are painted on, so the reader is looking at the same picture you are — a covered word is not in those pixels, and cannot come out of them.

What can it find?

Three things. Words you type, wherever they appear in the text. The same words where they appear as part of a picture — a logo, a screenshot, a scanned page — which an ordinary search cannot reach. And anything you pick out with a box: a signature, a stamp, a face, a letterhead. Pick one and every other copy of it in the document is found too, at any size.

There are also built-in detectors for the usual things that leak: email addresses, phone numbers, web addresses, dates of birth, and street addresses and postal codes — the last two written the way they are in Britain and America, and the way they are in Singapore, Hong Kong and Vietnam. Tick the ones you want; each shows how many it can see before you commit to it.

Names of people are on that list, but not in the way you might expect. Nothing here identifies a name by looking at it: a name is capitalised words, and so is half a deal document. It reads the anchor beside it instead — a job title under it, a "T:" or an email under it, a sign-off or a "Name:" above it, or a title beside it on the same line. That is a signature block, a contact block and a team page, which is where names actually sit, and it works for Vietnamese, Chinese and English names equally because it never reads the name.

Anything it misses, type into the box — a name typed once is then found in the text, in the lettering of a screenshot, and inside a logo, which no detector can do.

How does it read words inside pictures?

It reads the lettering on each page the way a scanner does. That happens in your browser too — the reader is about seven megabytes, fetched once from this site and then kept by your browser, and it is not sent your document.

No large language model is involved, here or anywhere else in the tool.

Nothing is perfect. What if it misses something?

Then you catch it, which is why nothing is covered until you press Redact and why every mark stays visible until then. You can draw your own boxes anywhere, and click any proposed mark to take it off.

There is also a second opinion: once a redaction is done, you can ask it to check every page again by shape rather than by reading. That catches words in unusual type that the reading stumbled over. Anything it finds is added in amber so you can see what is new, and it runs in the background while you carry on.

Why is my file bigger afterwards?

Because the new file is made of pictures rather than text, and pictures take more room than letters. That is the cost of the guarantee: the text cannot be recovered because there is no longer any text there.

There is a "lossless output" option if you would rather have a larger file with no compression softening.

What are the placeholders for?

Optional labels printed on the bars — [NAME_1], [EMAIL_2] — so a later reader can still follow a sentence that has had something taken out of it. They are off by default, because a plain black bar is the right output for most documents and a label tells the reader a little about what was removed.

If you turn them on you can download a key listing what each label stood for. Keep that separate from the redacted file: it undoes the redaction.

What happens if I close the tab?

Everything is gone. Nothing is stored, no account exists, and there is nothing to delete afterwards. The flip side is that a document you have not exported yet is lost too — so the tool warns you before you leave with one open.

Can I see the code?

Yes. It is free and open source: read it, run it yourself, or check what it actually does with your file.

Source on GitHub