Security
What we will not let happen. And what stops it.
minidesk is a multi-tenant system. The hard rule is NO CROSS-COMPANY LEAKS. EVER. The wall is built into the database, not asked of the AI. Below is exactly what that means, and the four rules that hold it up.
Rule 01
NO CROSS-COMPANY LEAKS. EVER.
Two companies on minidesk can share a phone number, an email, an account, a workspace — and the data they see is still completely separate. The wall is built into the database, not into a prompt.
Rule 02
AI scope = the asking human's membership scope.
What the assistant sees is exactly what the person asking it could see, no more. A clerk chasing a delivery cannot ask the assistant to surface the margin. The clerk's view is the AI's view, and only the clerk's view.
Rule 03
A plugin can never exceed its declared and granted scopes.
The plugin is the floor's tenant. It can read what it was granted, do what its manifest declared, and nothing else. Asking for more is asking the database to lie, and the database does not lie.
Rule 04
Every action is attributed.
To a person, a plugin, or an AI acting on a person's behalf. Every send, every write-back, every approval is logged. There is always an answer to 'who did this, under what authority?'
What we will not do
- We do not use your business data to train a shared model.
- We do not sell aggregated data to anyone, in any form.
- We do not run a third-party identity graph across our customers.
- We do not let the assistant break the wall even if it is asked nicely.
- We do not promise features that would force us to weaken any of the above.
The long version
The detailed security and data-processing documentation is on the roadmap. The full report — penetration test, infrastructure diagram, and the signed-off data-residency story — is shared with customers under NDA on request. Email security@minidesk.co.