Agents only get what you allow
Every agent runs inside permissions you approve, and ours follow the same rules as any builder's.
You see everything it can access before you install it
Before you install an agent, you see in plain words everything it can read. It gets nothing else.
Every request is checked, every time
Agents never reach your tools directly. Our gateway refuses anything an agent didn't declare or you didn't grant, and stops one stuck in a loop.
Nothing is sent until you approve it
Emails, Slack posts and CRM changes wait as drafts for you to edit, approve or reject. You can let an agent send on its own once it's earned it, and only to people and channels you list.
Every request is logged
Allowed or refused, every request is in your activity log with the reason. Agents can't read or change it.
Switch any agent off, at once
Turn an agent off and its next request is refused, even halfway through a run.
Your data stays yours
Walled off from other businesses
The database itself keeps each workspace's data apart from every other.
Keys stay on our servers
The keys to your tools never reach your browser, our logs or an agent.
Planted instructions go nowhere
An agent tricked by a planted message still can't go past what you granted. The gateway decides rather than the model.
Builders' agents are attacked first
Outside builders' agents pass attack tests and our review before they're listed.