Security
Powerful agents, firm guardrails
Four things the product enforces, and how each one works.
Workspaces are isolated at the database
Every table that holds agency data is scoped to a workspace, and the database enforces that with row-level security. The application connects with a role that cannot bypass it. Asking for another workspace’s record answers as if the record does not exist.
The ledger is append-only
The application can add journal entries but cannot edit or delete them, and the database refuses an entry whose debits and credits do not match. A mistake is corrected with a reversal that stays visible. The credit ledger that tracks agent spend works the same way.
Agent work waits for a person
Agents propose drafts, plans and suggestions. A run’s finished work waits for a person to approve it before it completes. Agents have no way to launch a campaign or send an email; those start with a person.
Agents hold no database or integration credentials
An agent acts only through a gateway that checks each call against what that agent is allowed to do. The agent service has no database connection and holds no integration tokens of its own.
This page describes how the product is built. It is not an audit report or a certification. To report a security concern, use the contact page.