Control over AI
Detection engine

65 data types. 11 categories. One engine.

BeeSensible recognises personal, financial, medical, and technical data in the apps your team uses. English and Dutch, on-device or in the EU, configurable per team.

Draft message

Hi team, Ingrid Lindqvist will be joining the project. You can reach [email] directly; her account is NL91 ABNA 0417 1643 00.

Warning
Names, emails, phone numbers
Critical
IBANs, BSNs, credit cards, API keys
Handled
Replaced before sending
65
Sensitive data types
11
Categories
~96%
F1 on our internal benchmark
Local or EU
Where detection runs
0 sec
Content stored
11 categories

Every type of sensitive data your team touches.

Detection profiles let admins toggle categories on and off per app. A healthcare profile enables medical codes; a finance profile enables IBAN and card data. Start from the default set or from blank, and adjust as you go.

Critical
Identifiers
15 types
  • · BSN (NL)
  • · IBAN
  • · Passport number
  • · Driver's licence
  • · Case number
Critical
Medical
12 types
  • · Medical code
  • · Diagnosis
  • · Medication and dose
  • · Health insurance ID
Standard
Location & address
8 types
  • · Street address
  • · Postal code
  • · City / Region
  • · GPS coordinates
Critical
Technical
8 types
  • · API key
  • · IP address
  • · Device ID
  • · Username
Critical
Financial
7 types
  • · Credit card number
  • · CVV and expiry date
  • · Crypto wallet
  • · Amounts
Standard
Demographic
5 types
  • · Gender
  • · Language
  • · Marital status
  • · Nationality
Standard
Contact
3 types
  • · Email address
  • · Phone number
  • · Web address (URL)
Standard
Travel
2 types
  • · Flight number
  • · Ticket number
Standard
Temporal
2 types
  • · Date of birth
  • · Event date
Standard
Person
2 types
  • · First name
  • · Last name
Standard
Organisation
1 types
  • · Company name

The severity shown above is just an example. You fine-tune exactly what counts as critical or standard to match your organisation's policy.

Detection profiles

Configure exactly what counts as sensitive, per app.

Every entity type has an on/off toggle and a severity level. Admins build profiles and assign them to apps; any app without its own profile keeps using the default. A strict profile for ChatGPT doesn't have to apply to LinkedIn.

  • · Toggle individual entity types without rebuilding a profile from scratch.
  • · See an example value next to each data type in the editor.
  • · Assign one profile per app, or reuse across your full stack.

Detection Profiles

Choose which sensitive data BeeSensible detects per app, and how strictly

What to detect
Choose which data types this profile detects, and how prominently each is flagged.
DetectLevel
Emaile.g. jan@bedrijf.nlStandardCritical
First namee.g. Sanne, DaanStandardCritical
Phone numbere.g. 06 12345678StandardCritical
IBANe.g. NL91 ABNA 0417 1643 00StandardCritical
BSNe.g. 111222333StandardCritical
How it works

Detection built for live typing, not batch scanning.

BeeSensible analyses text as the user writes. That happens on-device through the desktop app, or on BeeSensible's EU servers. On-device the text stays on the machine; in the cloud it is processed in working memory and discarded. No content log, no clipboard access.

On-device or in the EU

Text is checked as you type, never collected for later.

Detection runs on-device through the desktop app (macOS or Windows), which also hosts detection for the browser extension, or on BeeSensible's EU servers. On-device, the text never leaves the machine. In the cloud, it is processed in working memory and discarded. Nothing is stored or logged; only counts reach the dashboard.

Profile-scoped

Active entity types follow the detection profile.

Admins configure which types are active per app. A strict IBAN-heavy profile on ChatGPT does not have to apply to Slack; each app follows the profile you assign it, and the default profile covers the rest.

Two levels

Standard and critical get their own marking.

Critical types (BSN, IBANs, credit cards) get a red, high-visibility marking. Standard types (names, emails) get a yellow one. You set the level per data type in the profile.

Detection quality

Accurate enough to trust. Quiet enough to use daily.

Accuracy

Measured per entity type, not as a single average.

The detection is evaluated across every category and severity level. A strong overall result can't hide a weak class because each entity type is checked on its own.

Low noise

Built to avoid interrupting people unnecessarily.

Precision is tuned for each entity type. Common words like 'name' or generic numbers don't trigger warnings. Patterns need to match structure and context.

English and Dutch

Both languages, held to the same standard.

Dutch-specific types like BSN, Dutch IBANs, and KvK numbers are validated against Dutch-language samples. English entity types are validated against English-language samples. Neither is treated as a second-class case.

Continuously improved

Updated by BeeSensible, never trained on your text.

The detection is built and refined by the BeeSensible team. Whether it runs on-device or in the EU cloud, text typed by your users is processed in memory and discarded, so it never reaches a training dataset.

Roll detection out per app.

Pick the AI tools and email apps your team uses most, assign a profile to each one, then add the rest when you're ready.