Frequently asked questions.

We publish the numbers rather than adjectives. On 5,000 samples of the public ai4privacy/pii-masking-200k dataset, Hexagone AI detected and masked 96.95% of ground-truth personal data entities against 75.30% for Microsoft Presidio. On an augmented version of the same set containing typos, abbreviations and paraphrases, the figures were 93.34% against 65.89%. On RAT-Bench, the re-identification benchmark from Imperial College London, Hexagone AI recorded the lowest re-identification rate of the systems compared, 14% at level 1 and 27% at level 2, against 22% and 33% for Azure and 26% and 58% for a GPT-4.1 masking prompt, with no direct identifier recovered. The full report and methodology are at hexagone.ai/research, with no form to fill in.

Presidio is an open-source Python library, so you keep control but you build and maintain everything around it: document handling, OCR, cross-document consistency, audit trail and interface. Cloud services such as Azure AI Language are well-supported but require uploading the document, which is the one thing professional secrecy forbids, and they tend to over-redact: Azure scored 0.55 BLEU on RAT-Bench against our 0.83, meaning much more of the text is destroyed than needs to be. The full criteria-by-criteria comparison is at hexagone.ai/comparatif.

PDF, Word (.docx), Excel (.xlsx), PowerPoint (.pptx) and plain text. The app rebuilds the document with the same layout and formatting, just with the personal data replaced. Scanned and image-only PDFs are handled by local OCR, so a photocopied case file works the same way a digital one does.

A redacted document hides information but breaks the flow of the text, and the operation is irreversible. Hexagone AI replaces names, addresses, IDs and amounts with realistic placeholders, so the document is still readable and usable for analysis, summarisation or sharing with an AI assistant. Because the mapping is kept on your machine, you can also restore the real identities in whatever the AI sends back.

Yes, across a whole folder. This matters more than it sounds: without cross-document consistency an AI cannot follow who did what across a case file, and the answers it gives become useless. Point Hexagone AI at a directory and every occurrence of the same person, company or reference gets the same substitute in every file.

Only the first time, to download the app and the local AI models. After that you can disconnect from the internet and keep working. Updates and the optional chat features need a connection, but masking itself is fully offline.

Yes, that is exactly who we built it for. Because the data never leaves your machine, there is no transfer to a third party and no processing agreement to negotiate. Every detection and substitution is logged locally and exportable as CSV, which is what a DPO or a regulator will ask for. We provide documentation you can hand to your compliance team, and you can email us at contact@hexagone.ai for the legal pack.

Yes, on the Enterprise plan. It deploys inside your own infrastructure, including fully air-gapped environments with no outbound network at all, with SSO, a command-line interface, dedicated onboarding and a custom SLA. Write to contact@hexagone.ai and we will scope it with your security team.

Yes. The free trial gives you unlimited processing, unlimited chats and unlimited AI workspace usage for one week, with no account or credit card required.

Hexagone AI Desktop is priced per user. Starter is €49 per month billed annually, or €59 monthly. Pro is €79 per month billed annually, or €99 monthly, and adds unlimited pages plus anonymization inside the AI workspace. For enterprise deals with many licences, contact us directly at contact@hexagone.ai. Full details at hexagone.ai/tarifs.

Most likely yes. The app is built for everyday work laptops, not gaming rigs. We support macOS Apple Silicon and Windows 10/11 64-bit, with 8 GB of memory minimum. See the details at hexagone.ai/get-started.

Yes. You can enable or disable categories (people, organisations, addresses, phone numbers, IDs, financials, etc.), define custom entity types specific to your practice, and set per-document overrides that persist between runs. Every detection appears in the preview pane and you can edit anything before exporting.

Originals stay where you put them. Masked outputs are written next to the source by default, in a location you choose. Hexagone AI Desktop does not keep its own private copy of your documents.

Email contact@hexagone.ai. We are a small team of founders, you will get a real human reply quickly.