Frequently asked questions

Can the quality of the produced pages be set?: Yes. Pages come out as images: three levels are offered, light (150 dots per inch), normal (200) and fine (300). Normal is enough to read on screen and to print; fine roughly doubles the file size, and is for a document that will be examined closely.
What does it cost?: One document at a time is free on this site, and will stay free: no account, no page limit, no watermark. During the launch, the whole tool is open, batch processing included. Afterwards, work in series, team use, self-hosting and the SDK will be paid plans: see Pricing.
Is my document sent anywhere?: No. It is read by your browser and processed in the tab. You can check this by cutting the network once the page has loaded: the tool goes through to the end. See the Privacy page.
Is the redacted text really removed?: Yes. The file produced is a new document: each page in it is an image on which the black boxes are painted, with searchable text rewritten without what is redacted. Before offering it to you, the tool reads it back and searches it for each redacted item.
Does the document become anonymous?: Not necessarily. Klarmask redacts and pseudonymises; a document can remain recognisable through its context. To anonymise a document, in the GDPR sense, nobody must be able to identify the person any more, even by cross-referencing information. See pseudonymisation or anonymisation.
Why do the pages become images?: Because it is the only safe way to remove the text under an area: nothing from the original page is copied. The kept text is rewritten on top, invisible, so that the document stays searchable. The page loses its links, its forms and its accessibility tags.
Are names detected?: Yes, by a model that runs in your browser. It is downloaded once from this site (134 MB for the light version), verified by its fingerprint, then stays on your computer. It suggests the names of people and companies, and groups the forms of the same person (“John Smith”, “Mr Smith”, “SMITH John”). It is trained on French and can miss names, especially in other languages: “Redact everywhere” redacts a missed name throughout the document.
What about Word documents, scans and photos?: Word documents (.docx) are supported: text, tables, headers, footers, notes and text boxes, with tracked changes accepted and comments removed. Scans and photos of documents (JPEG, PNG, WebP) are read by text recognition, in your browser, then processed in the same way. Handwriting or a blurred scan reads poorly: reread the image, and draw an area over anything that is not suggested.
What is an item “to review”?: A number that has the right format but whose check digit does not add up. It is often a typo in the document: the data is still personal, so the tool shows it to you.
Can I re-identify after pseudonymisation?: Yes, if you planned for it. After a pseudonymisation, the tool offers to save the mapping table (“[Person A] = John Smith”), encrypted in your browser with a passphrase you choose. To read it again, use “Open a mapping table”, below the drop area: the file and its passphrase are enough. Klarmask knows neither; a forgotten passphrase cannot be recovered. As long as the table exists, the document is still personal data. The same table can also be reused for new documents (same labels for the same people), or restore the names in a pseudonymised document that comes back to you.
Are signatures, stamps and faces redacted?: They are suggested. Each page is read again as an image: signatures, stamps, faces seen from the front and QR codes or barcodes found outside the text appear as areas to decide, like names. Anything missed (a face in profile, a faint signature) is covered by drawing an area.
Can I keep a record of what was redacted?: Yes: the redaction report, offered with the file produced. It states what was redacted by type and by page, and the fingerprint of each file produced, with no redacted value.
Can I produce several versions of the same document?: Yes. Add a version in “Several versions”, tick the types it leaves visible, and both versions come out together, each with its report. Labels are the same in all of them.
Can I interrupt a review and resume it tomorrow?: Yes. “Continue this review later” saves your decisions in a file encrypted with a passphrase. Reopen the same documents and read that file back: the tool checks each document's fingerprint before resuming.
Can I process several documents at once?: Yes. Drop them together: they are read as a single case file. The same person is a single suggestion, decided once, and carries the same label in each document. Each document is rebuilt and checked separately, then everything is saved in a ZIP archive.
Where are my rules kept?: In this browser, on this computer, and nowhere else: the list of names or words to always redact does not leave the device. It is erased with the site data, and in private browsing it does not last beyond the session.
Which browsers?: Chrome, Edge and recent browsers. Firefox and Safari have not been checked yet.
Do I need an account?: No. No account, no cookie, no sign-up.
Does the tool work offline?: Yes. After a first visit to the tool page, the browser keeps its pages and code: it reopens and works without a network, from dropping the file to saving it. The name model, if it has been downloaded, is stored on the computer; the scan reading engine is kept after its first use.
What do you count?: The number of documents redacted and checked, and their number of pages, for the counter on the home page. One request when you save, to this site, with nothing that identifies you and nothing from the document. The details are on the Privacy page.
Who is responsible for the shared document?: You are. The tool suggests and carries out; the decision on what is redacted, and the sharing, remain yours. See the terms of use.