Realtime Privacy: detection while people type
What the Realtime Privacy module does, where it works, and what you control as an admin.
Hi team, Ingrid Lindqvist will be joining the project. You can reach [email] directly; her account is NL91 ABNA 0417 1643 00.
Realtime Privacy flags sensitive data at the moment someone types or pastes it, before the message goes out. It is the layer your team sees every day: a highlight under a name, an account number, or a passport number in an AI chat, an email, or a chat message, with the choice of what to do about it.
What your team sees
While you type, BeeSensible checks the text and highlights what is sensitive: yellow for regular sensitive data, red for critical data. For each highlight the person decides: replace it with a realistic stand-in, mask it, or remove it. BeeSensible never blocks sending and never changes text on its own. Ignoring a warning and sending anyway is allowed; it is simply counted on the Insights screen.
Detection runs in two places, on the same engine:
- The browser extension (Chrome and Edge) works in AI tools, email, chat, and social platforms.
- The desktop app (macOS and Windows) adds four apps the browser cannot reach: Outlook desktop (classic and new), the ChatGPT desktop app, the Claude desktop app, and Microsoft Copilot. There, the chosen fix is applied directly in the app's own text field.
What you control
- Detection profiles decide which of the 65 data types are highlighted, at which level, and per app. The profile for external AI tools does not have to apply to an internal chat. See Detection profiles, categories, and levels.
- Where detection runs: on BeeSensible's own servers at European providers. See The detection engine.
What you see back
Activity lands in the dashboard under Insights: counts and patterns per period, per app, and per data type, plus how detections were handled. Never the text itself, never a per-employee view. See Reading the Insights screen.
Where the text goes
The text goes to BeeSensible's own servers at European providers: Scaleway in Amsterdam, the models at Hetzner in Germany. There it is checked in working memory and discarded straight after. Nothing is stored as content, and no external AI service is involved.
