The problem nobody wants to write down
Your team is already pasting confidential data into ChatGPT.
Client records, invoices, contracts, ID numbers. You know it's happening. This does the same job — without the confidential parts ever leaving the device.
Try it. Nothing here reaches a server.
Type or edit the text below. The right panel is exactly what an AI provider would receive. Open your network tab if you like — there are no requests.
Detection uses checksums, not guesswork: a national ID must satisfy its control letter, an IBAN must pass mod-97, a card must pass Luhn. The substitution map never leaves this tab — which means we cannot reverse it, even if we wanted to.
How it works
- Redacted before it leavesIdentifiers are replaced with placeholders in your browser, before any request is made.
- The model answers on placeholdersIt reasons over
[ID_1]and[IBAN_1]— never the real values. - Restored only for youYour browser puts the real values back into the answer. The server never held the key.
Never used for training
Requests are routed only to providers that contractually do not train on submitted data. Providers that do are excluded unless you explicitly opt in.
Documents too
PDFs, Word, Excel, scans and images are converted and searched without the file reaching a third party.
A processing agreement, not a promise
A GDPR Article 28 data processing agreement, a published sub-processor list, and EU-hosted options.
Ask anything, in any language
Chat, document search, transcription and voice — the parts you already expect, with the parts you don't.
Worth 15 minutes?
Tell us what your team handles and we'll show you exactly what would and wouldn't leave your network. No slides.
Email usWe reply to every message. If it isn't a fit we'll say so.