The web app.
From file to restoration.

On first use, the guide offers Download. The model is the component that detects personal data. Select Download and wait for downloading, verification and startup. You need a connection to receive about 343 MB. The time depends on your device and connection.

During download, the button is disabled and a message shows progress. Wait for “Model ready” before continuing. If an error appears, check your connection and available storage, then retry Download. If your browser does not retain the model, a message explains this and another download may be needed. Skipping the guide does not remove the model requirement: analysis will request it.

Complete or close the initial guide. Open a file using File, or use the text area. Use fictional data for your first test. PDFs, DOCX files, images and text follow different workflows: native PDF text is read directly; scans and images require OCR, which reads text from an image.

Importing a file may start analysis. Use Analyse to start or repeat it after relevant changes. Wait for the result and check that every page is present. A ready model does not mean that your document has already been analysed.

Review the entire document and detected-data list. Correct values, exclude unnecessary selections and add missed data. Check signatures, images, dates, locations and details that make the case recognisable. Manual areas can cover items that detection did not select.

In the Export menu, choose With reversible codes or Blocks only. A block covers content; that block alone cannot reconstruct it. A code replaces a value and is linked to a mapping. Export produces the format selected in settings. PDFs are rebuilt with redactions applied to the pages; keep the original separately.

Select TXT with codes to export page text using replacements from the same mapping. This file supports text work with AI. It does not preserve images or the complete original layout. In PDFs, text intersecting a manual redaction is excluded: check that necessary passages are not missing. Open the result before sending it.

Open Mapping and select Save. The mapping links each code to its original value and contains sensitive information. Keep it separate from the TXT you share. Check the saved format and any protection: not every workflow produces an encrypted mapping. Do not send the mapping to an AI service.

Choose a service permitted for your work and send only the necessary content. Ask: “Preserve every code exactly as written; do not change, invent or remove codes.” The external chat receives whatever you choose to send. Remaining details may still allow identification.

Open Restore. Load the same mapping, then open the AI response TXT or paste the response into the input area. The result replaces codes known to the mapping. It cannot reconstruct values deleted with a block or correct facts written by AI. If AI changes or removes a code, review the response against the source.

Select Download restored TXT and open the file. The command stays disabled if no mapping code was found. Check names, roles, unrestored codes and the response itself. Protect the restored text like the original document: it contains the real values again.

Content cover that does not permit recovery through the block alone.

Replacement for an original value, linked to a mapping.

Correspondence between codes and originals; store it separately.

Replacement of codes known to the mapping, followed by review of the result.