See exactly where personal data goes, what is encrypted, and what is not. Written for data protection and security officers.
Book a demoRequest the security pack
Get started
Precise on purpose: including the parts that are not covered.
Set once for the organisation. Every message: block it, or (from Pro) replace values with placeholders that are restored in the answer. One last check before an outside AI: ask the person, hide the value, or do not send. From Pro, the Web Search Guard also holds personal data back from outbound web search and external tools.
Detection happens on your hardware. The guard runs in a container you operate, on CPU, and nothing is sent to a third party to detect personal data. It looks for 21 categories, with checksum checks such as BSN, IBAN and Luhn. It exists only when the guard service is installed: a core-only install does no detection.
Only to the provider you chose. The request goes to that model provider, with placeholders or hidden values where your rules say so. If the guard is installed but unreachable, chat messages are blocked (fail closed). Tool results, memory and knowledge base passages are passed on unscanned, and large attachments are cut off unless set to fail closed.
Places the shield does not check: the swarm tier in direct chat, typed messages in template chat and the webpage builder, voice transcripts and the learning coach.
Content encryption is off until an admin switches it on. With managed keys (from Pro), an organisation root key holds each person's key, so a password can be reset and the operator can read the data. With zero-knowledge (Enterprise), only the person's own secret unwraps the key: no master key, no back door. Even then, chats shared into a project, meeting transcripts and memories stay readable to the operator, and knowledge base documents are not encrypted by the app. TLS ends at the reverse proxy you run.
Get started
Open the settings and choose what happens to personal data. Compare the actions.
Try the Privacy Shield
Sample data. Nothing you do here is saved.
Who gets in, who did what, and how you measure it.
Use the directory you already run. Single sign-on with Google and Microsoft Entra ID (from Starter), Nextcloud sign-in, and TOTP two-factor authentication with recovery codes.
Access that follows people's jobs. Groups sync from Microsoft Entra ID (on demand, or on a schedule once you switch that on) and from Nextcloud. Six built-in roles, including Data Protection Officer and ISMS auditor, plus custom roles.
See what happened and what it cost. Every integration call is logged with its host, the company that runs it, its country, status and duration, without content. A usage overview shows what was used. Audit log export is part of Enterprise.
The Compliance Center (Enterprise, generally available) runs 84 readiness checks across 10 frameworks, including GDPR, ISO 27001 and the AI Act: 60 automated, 14 hybrid and 10 attested. Each one reads the configuration of your own install and tells you how to fix what fails. Readiness checks, not a certificate.
Get started
We hold no ISO 27001 or SOC 2 certificate, and we have not had an external penetration test yet. The limits of the Privacy Shield and of encryption are listed above, next to what they do.
Tell us what you have to prove and where your data has to live. We will show you how Bee Flow fits.
Request the security pack
Book a demo