The small thing you didn't check, and what it costs.
Six five-second checks nobody does, standing between an ordinary file and an expensive, public mistake. Everything runs in your browser: nothing you drop in is ever uploaded, because the file is the thing you were trying to protect.
No accounts · no uploads · no trackingNot sure which one?
Pick what you are about to send.
The six tools
Shade.
LiveScreenshots, ID documents and photos you're about to post. Finds what's sensitive and destroys the pixels: no blur, no pixelation, because those are reversible.
Mimir.
BetaChecks every reference against seven real databases. Flags the ones that don't exist, the ones whose details don't match, and the ones that have been retracted.
Argus.
LiveFinds text a document hides from you but not from an AI reading it: white-on-white, sub-pixel fonts, invisible Unicode, direction overrides, smuggled instructions.
Kelpie.
LivePaste a URL and see what's riding along: tracking IDs, your own email address, session tokens, and whether that shared document link hands over edit access.
Selkie.
LiveWhat your spreadsheet is still carrying: hidden sheets, hidden rows and columns, cached pivot data from deleted sources, comments, and the author's name.
Gryphon.
LiveDrop the zip before you send it. Finds .env files, .git history, private
keys and hardcoded credentials: with the secrets shown redacted, never printed back.
Why these run locally
The file is the secret
Every tool here handles something you specifically didn't want to send anywhere. Uploading it to check it defeats the check.
An assistant can't replace them
Asking an AI means handing over the exact thing you're protecting, and a model can't verify a DOI exists. That takes a database lookup.
Nothing to opt out of
No account, no cookie banner, no analytics. Close the tab and there's nothing left behind, here or anywhere else.
Questions
Are my files really not uploaded?
Yes. Every tool reads your file with the browser's own APIs and processes it in the page. There is no server, no database and no storage bucket anywhere in this project. You can verify it: open your browser's network tab and drop a file in: you will see no request carrying it. The one exception is Mimir, which looks up reference metadata in public citation databases; it sends the citation text, never your document.
Is it free, and what is the catch?
Free, with no account and no limits. The site is funded by contextual advertising: ads chosen by the page's subject, not by anything about you. They run inside a sandboxed frame that cannot read the page, so they never see the files you drop in.
Why not just ask an AI assistant to check the file?
Because that means uploading the exact thing you were trying to keep private. It also does not work for the hard cases: a language model cannot confirm a DOI resolves, cannot recover a spreadsheet's hidden pivot cache, and will not reliably notice white-on-white text that its own reader silently absorbed.
Does redaction here actually remove the data?
Yes, and that is the point. Shade paints opaque rectangles and then re-encodes the image, so the covered pixels no longer exist in the file. It never uses blur or pixelation, both of which are routinely reversible: that is precisely how the aCropalypse screenshots were recovered.
Do these work on a phone?
Yes. Every tool works in a mobile browser. Shade's OCR downloads a language model on first use, so the first run on mobile data is a larger download than the others: after that it is cached.
What do the tools not catch?
Each tool states its own limits on its page, and the project README lists every known gap. In short: Gryphon does not open a zip inside a zip, Selkie is deliberately noisy on heavily filtered workbooks, and no tool here will ever tell you a file is "safe": only what it checked and what it found.