Safe GPT that can’t see your personal data
Data is hidden automatically before it reaches the model. Works with text and DOCX/PDF files.
Like ChatGPT — without your data leaving
You type as usual. SecureChat tokenizes PII before the model; on your screen the reply is decoded back.
Files — automatic or manual
The same protection inside the product. Not only for AI: clean a contract for a client too.
Attached in chat — personal data is masked on the fly, before the model.
What we catch
Local formats from day one — not generic templates.
14 digits, with or without spaces
Including Cyrillic series
Uzcard / Humo in any written form
STIR / TIN for orgs and people
+998 and local formats
Latin and Cyrillic
How we treat data
- Logs are kept for a limited period, for diagnostics only.
- Nothing is used to train models.
- The server is on Uzbekistan territory.
Common questions
Is it safe to paste documents with personal data into ChatGPT?
Technically you can, but you are sending them to a company outside Uzbekistan. Personal data legislation requires citizens’ data to be processed and stored on servers inside the country, and public AI services do not do that. SecureChat removes the personal data from the request before it reaches the model.
What does the model actually see?
Text with markers in place of the personal values: {{Person_1}}, {{Pinfl_1}}. The values stay on your side and are put back into the reply on your screen.
What does SecureChat detect?
PINFL, passport series and numbers including the Cyrillic АА and АВ forms, STIR, card and bank account numbers, +998 phone formats, and names in Latin and Cyrillic — including “o‘g‘li” and “qizi” forms and initialled records like “Kenjayev O. O.”. The model distinguishes more than forty data types in all: addresses, dates of birth, contract numbers, amounts.
How is this different from deleting the data by hand?
By hand you forget. In a ten-page contract the PINFL turns up in the header, in the payment details and in the annex — missing one is enough. And it costs you time every single time.
Where is the data stored?
Nothing but the markers reaches the model. Request logs stay on our side — they are needed for diagnostics and statistics, and are kept for a limited period. We do not train models on them. Exact retention periods and server locations are set out in the contract for business customers.
When does access open?
We open access in batches and message you on Telegram when your turn comes. The more clearly you describe what you need it for, the sooner we get back to you.
Early access
For B2B partnership, apply at secureprompt.tech