Frequently asked questions
The questions we get first.
If you still have doubts or questions, you can write to us directly and we will reply within 24 hours.
Confidentiality
Do filings pass through a server of yours?
No, and there is no server they could pass through. The analysis runs on your computer, language model included. The app opens two addresses only, updates and your account, and every request to any other one is cancelled before it goes out.
Does it work without a network?
Yes, in full. Open the app, open a matter, analyse it, export it: none of that asks for the network. It is the test we redo at every release, because a lawyer in a courtroom with no signal opens this app.
Do you collect usage statistics?
Not in the program: no analytics and no automatic error reporting, so what you do on a filing never leaves your computer. On this site, statistics only start if you accept them in the banner, and no counts as much as yes: the page works the same either way.
What does the update check see?
An IP address and a user agent, like any download. The request is a GET on a static file that is the same for everyone: it carries no email of yours, no token, no identifier for this computer. The flip side is that we do not know how many copies are installed.
Where does the mapping between placeholders and real names end up?
On your computer, encrypted with the system keychain. If you delete the matter, the mapping goes with it and from that moment the AI’s answer can no longer be reassembled. That is deliberate: a dictionary that outlives the matter is an archive of personal data nobody asked for.
The product
What is the difference between redacting and anonymising?
Redacting means covering: nothing is left under the black box, and nothing comes back. Omissia substitutes instead, and a substitution can be undone. It puts a stable placeholder where the name was, you paste the text into the model, and when the answer comes back the app puts the real values where they belonged. That is why it exists: you can file a redacted document, but you cannot read an AI answer back onto it. What we do is technically pseudonymisation rather than anonymisation, and we write that down because it is the difference that matters the day you have to explain what you did.
Which files does it read?
.pdf, .docx, .odt, plain text (.txt, .text, .md, .markdown) and the images of a scan (.jpg, .jpeg, .png, .tif, .tiff), plus pasted text. A PDF with no text layer is no longer refused: the pages that do not carry their text are read here on your computer, and an image becomes a one-page PDF the moment it comes in. No script in the document is ever executed. A file renamed to .docx is refused, not opened.
Does it read scanned PDFs?
Yes, and the pages stay on this computer. A PDF with no text layer is no longer refused: the app recognises only the pages that do not carry their text (the ones that already do cost nothing), says which page it is on, and stops with Cancel. 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.
Is it only for the AI?
No. The model is the occasion most people install it for, not the only place a filing ends up: the colleague who has to give an opinion, the trainee drafting the brief, the technical expert, the client asking to see a precedent close to their own, the other side receiving an attachment. The round is the same and it stops one step earlier: you export and never use the way back. On a PDF the values are not covered but removed from the file, along with the metadata and the properties. When you do not need the way back at all, the case opens without a dictionary and no mapping is built or saved.
Which systems does it run on?
macOS, Windows and Linux. On macOS and Windows it updates itself; on Linux it does not, and the Download page explains why: an AppImage is not signed, and an update that verifies itself has verified nothing.
How reliable is the detection?
Reliable enough to spare you the hunt, never enough to replace the re-read. Values with a checksum (tax code, VAT number, IBAN, cards) are confirmed by an algorithm. Person names are not: a model proposes them, and a proposal can be wrong. That is why the app replaces nothing you have not seen, and why it tells you how many values are still in the clear before you export.
Can I add terms of my own?
Yes. Names of opposing parties, internal abbreviations, company names: they go into the matter’s dictionary and are found like everything else.
Which languages is it in?
The app speaks Italian, English, Slovene and German, menu bar and native dialogs included. This site, for now, Italian and English.
Licence and buying
Do I need an account to use it?
Yes. The program is used with an account: signing up starts a three-day trial with every feature, and when it ends you need to buy.
Do you charge per page?
No. You pay for the seat, and you use as many pages as you need. A four-hundred-page file costs the same as a four-page one, and there is no page pack to buy again halfway through the month.
What happens when the licence expires?
Without an active licence the program cannot be used: when the three-day trial ends you buy, and when a licence expires you renew to keep going.
How do I buy it?
On your own, from the account area: you create an account, try it for three days, and when the trial ends you buy directly from there, where the invoices live too. From five seats up you write to us and the price is agreed together.
Still a question left?
Write to us. Someone who wrote this product answers, not a form.