Privacy before the prompt

Use any AI.
Keep your real data.

Mask finds the sensitive parts of your text - names, emails, phone numbers, cards, account numbers, addresses, amounts - and swaps them for tokens before anything reaches an AI model. When the answer comes back, it puts the real values back. The mapping never leaves your browser.

Mask your first text See how it works

Runs in your browser. No account. No upload. No storage. Detection coverage on our open test set of 12 synthetic business texts: 26 of 26 seeded values (100%).

Three steps between you and the model

You keep the keys. The model only ever sees the masked version.

01 - Mask

Real values out, tokens in

Paste your text. Mask detects personal and business data and replaces each value with a token like [PERSON_1] or [AMOUNT_2]. The token table is built in your browser tab and stays there.

02 - Ask

Send only the masked text

Copy the masked text (or a JSON envelope) into ChatGPT, Claude, Gemini, a free model, or your own pipeline. Or press "Ask the AI" and our relay sends the masked text to a free model for you - after a second check that nothing raw slipped through.

03 - Restore

Real values back in the answer

Paste the AI's reply into Mask. Every token is swapped back to its real value. To anyone watching the wire, the whole exchange was about placeholders.

What Mask catches

Detectors tuned for business text: client emails, payment instructions, HR notes, support tickets, legal drafts.

People names in context and common first names Organisations companies with Ltd, Inc, Bank and more Emails and phone numbers in international, Ghana, UK and US formats Payment cards verified with the Luhn check Account numbers IBAN, UK sort code + account Government IDs Ghana card, US SSN, UK NI Addresses street addresses, UK postcodes, US state + ZIP Amounts $, £, €, ₵, GHS, NGN and currency words Dates of birth after DOB or "born" cues Your own terms project codenames, client names, anything you add

What Mask does not promise

We would rather lose a sale than oversell a privacy tool. Read this before you rely on it.

Masking cuts exposure. It is not anonymisation. The meaning that remains in your text can still say a lot about you, and a model provider can store or learn from whatever you send it. Mask removes the specific values it detects - nothing more.
No detector catches everything. Mask measured 26 of 26 seeded values (100%) on its own published test set of 12 synthetic business texts. Your text is different. Read the masked version before you send it.
Context can re-identify. "The CEO who signed the Accra port deal" needs no name to be identified. Add distinctive phrases as custom terms when it matters.
The demo model is a third party. "Ask the AI" sends masked text to a free external model. That is the point of the product - but if a text is so sensitive that even a leak in the remaining meaning would hurt, do not send it to any model.

Pricing

Try it free right now. Pay when it becomes part of how you work.

Desk

free forever

Mask and restore in your browser
1,000 characters per text
JSON envelope export
Open test set and coverage report

Start free

Pro

per month

Unlimited text length
Custom term lists, saved locally
"Ask the AI" relay with pre-send leak check
Priority detection updates
Access code by email after checkout

Get Pro

Team

$29

per month, 5 seats

Everything in Pro
Shared custom dictionaries
Deploy-in-your-own-stack option
Email support

Talk to us

Questions

Where does my text go when I mask it?

Nowhere. Masking runs in JavaScript in your browser tab. There is no upload, no account, no database. Close the tab and the mapping is gone (Pro can keep it in your browser's local storage, still never on our servers).

What exactly is sent to an AI model?

Only the masked text, and only when you choose to send it - either by pasting it into an AI tool yourself, or through "Ask the AI", which runs a second leak check before sending and tells you which free model answered.

Does this make my data GDPR-safe?

No single tool does that. Masking is pseudonymisation: it reduces risk and supports a privacy-by-design workflow, but whether a dataset is anonymised depends on everything that remains in it. Talk to your counsel for a legal call.

How is this different from enterprise redaction platforms?

They are built for compliance teams with budget. Mask is built for the person about to paste client text into a chatbot: fast, browser-local, honest about limits, and priced for one desk.

Use any AI.Keep your real data.