Twelve euro fifty a month for the seat the filings live on.

Twenty-three types of personal data: names, tax codes, IBANs, licence plates, land registry data, docket numbers. The analysis runs on your computer, language model included, and values with a check digit arrive already verified.

A PDF with no text layer is not refused: only the pages that do not carry their text are recognised on this computer, with the page count and Cancel in plain sight. A .jpg or a .png of a scan comes in too: it becomes a one-page PDF on the way in. No image leaves the machine.

Each detection is shown where it sits in the document, with the sentence around it: it is the only way to decide whether “Milan” is the client’s residence or the name of the court. Nothing gets replaced without you seeing it. And what the recognition cannot catch (a signature, a stamp, a name a scan read crooked) you cover with a rectangle drawn by hand, which removes from the file whatever sits under it.

The text goes out with stable placeholders: [FULLNAME_1] stays the same person through the whole document, so the model still understands who does what. Before saving, the app counts how many values are still in the clear.

A tab in Word’s ribbon, two entries in LibreOffice’s Tools menu, three entries in Pages’ Services menu, with a shortcut you assign yourself. The review stays the app’s own, and the file comes back with no revisions, no comments and no properties carrying a name.

The app can accept requests from the other programs on this computer: twelve addresses on 127.0.0.1, the same analysis the window does, and a journal showing what knocked. It starts off, and you are the one who turns it on.

You paste the AI’s answer and it comes back with the names in place. The mapping between placeholders and real values never leaves your computer: it sits encrypted in the system keychain. If what you have is a file and not a text, “Open a document…” reads the placeholders back out of an already-anonymised PDF: there is no key inside the document.

Names of opposing parties, internal abbreviations, company names: the terms you add get found like everything else, in every document of the same matter.

On macOS and Windows they arrive by themselves. On Linux you download them from the Download page, and the reason is written there.

Italian, English, Slovene and German, menu bar and native dialogs included.

Someone who wrote the program answers.

A brief to summarise, a pleading to start the next one from. Parties, addresses and docket numbers become placeholders before the text reaches the AI, and the answer comes back with the names in place.

Board minutes, a contract, a report you want explained. Tax codes, VAT numbers, IBANs and amounts disappear first, and the client stays a placeholder even for the model.

A confidential agreement to reread, a draft you want torn apart. Company names, figures and bank details never leave the office: the AI gets the text with placeholders, you get the answer with the real names.