A pseudonymization service for small and medium businesses: names, clients, projects, places and identifiers are replaced automatically before a prompt reaches Claude or ChatGPT, and restored in the answer.
Early access — in development. Hosted in France and Europe.
What it does
Designed for teams who paste client data, contracts or HR files into AI assistants.
Detection runs in the browser, on the user’s computer: the original text is not sent to Observable.
Names, places and identifiers become tokens such as [PERSON_1]; the AI’s answer is shown with the original values restored, on the user’s screen only.
French identifiers checked by their control keys (SIREN, SIRET, VAT number, social security number, IBAN), phone numbers, emails and dates, plus a dictionary of your own clients and projects.
A dashboard for the data protection officer shows how many items were masked, by category and by user — never the text itself.
How it works
Employees keep using Claude and ChatGPT in the browser.
Before sending, a browser extension swaps detected items for tokens.
Tokens in the reply are turned back into the original values on screen.
Early access — in development
The pseudonymization API and the browser extensions for Claude and ChatGPT are in development: nothing is publicly available yet, and sign-in is for early-access accounts.
Pseudonymized data remains personal data under the GDPR. The service reduces exposure; it is not anonymization, and no detector finds everything. By design, Observable does not store the original text.