Safety

Trust by Design

Delegation only works when the delegate is reliable. Joined uses verifiable controls, receipts, and reversibility instead of blind trust.

All Features

Trust is not a feature.
It is the architecture.

Most AI products ask for trust. Joined is designed so trust can be audited.

Action logVerified
  • Sent follow-up to Sarah (approved)
  • Booked dentist appointment (approved)
  • Drafted cancellation request (awaiting approval)
Every action is reversible.

Six Principles of Trust

Trust in Joined comes from constraints baked into the system.

Transparent Actions

Every action is logged with what happened, why it happened, and what triggered it.

Explicit Permissions

Capabilities are granted one-by-one and can be revoked at any time.

Undo Everything

Mistakes are reversible by default, not permanent by accident.

Private Vault

Context stays encrypted, isolated, and under your control.

Full Audit Trail

Every skill, action, and decision is attributable and inspectable.

Consent-First Execution

Joined proposes first and executes only within your approved scopes.

Permission Model

You start read-only and graduate permissions as confidence grows.

Read
Remembers context but cannot act on it.
Draft
Prepares actions for approval.
Act
Executes inside scoped permissions with receipts.
Automate
Handles recurring tasks you explicitly approved.
Your Permissions
Send MessagesEnabled
Book Calendar EventsEnabled
Make PurchasesDisabled
Access Health RecordsDisabled
Legal Transparency

Trust claims backed by readable legal docs.

Joined’s trust model is mirrored in policy: explicit permissions, scoped data use, 30-day default telemetry retention, and a 7-day active-system purge target for deletions (with legal-hold exceptions).

Delegate With Confidence.

Start with read-only memory, then expand permissions as trust is proven.

Join the Waitlist

Your Agent Is Almost Ready.

We're onboarding in waves. Drop your email and we'll let you know when it's your turn.

No spam. Unsubscribe anytime.