Your agent holds your whole world.
Here’s why that’s safe.
The more it knows, the more it can do — and the more it must be auditable, reversible, and yours. Joined is built so trust can be checked, not just asked for.
You’re not handing it your grocery list. You’re handing it your inbox, your calendar, your board context, and the logins to run on your behalf.
That’s exactly why the controls below are the architecture, not a footnote. An agent this close to your work has to be the most accountable software you own.
Six principles, baked in.
Transparent actions
Every action shows what it did, why, and what triggered it — before and after.
Explicit permissions
Capabilities are granted one by one — and revoked the moment you want.
Reversible by default
Every action lands in an undo log you can roll back from — never permanent by accident.
Private vault
Your context is encrypted, isolated, and under your control. Never training data.
Full audit trail
Every skill, action, and decision is attributable and inspectable.
Consent-first execution
Proposes first, and executes only within the scopes you approved.
You start read-only.
Trust is earned a rung at a time.
Proactive about what to do, permissioned about what it executes. You move it up the ladder as it proves itself — and you can move it back down instantly.
Joined acts for you — without going rogue.
Watches
Reads the context you connect — nothing you didn’t.
Plans
Drafts the move and shows you, before anything happens.
Executes
Only inside approved scopes — and you can audit every step.
Trust claims, backed by readable docs.
Encrypted at rest, isolated per user, and access-logged end to end. Explicit, scoped permissions. 30-day telemetry retention. Deletions purged from active systems within 7 days, legal holds excepted. Your data is never sold and never used to train models.
Before you connect a thing, we’ll walk you through exactly how it’s stored, scoped, and deleted — trust you can check, not claims you have to take on faith.