Product
You approve the exact thing. Nothing else.
An approval is tied to the payload it was shown. Alter one character and Brigham.ai asks again.
Approval is a hash, not a vibe.
Brigham.ai fingerprints the exact action. If the payload changes after you approve, the approval no longer applies.
How an approval runs.
Propose
The agent submits an action. The tool pipeline holds it.
Notify
You get it in the app and as a notification.
Review
You see the exact payload, not a summary of it.
Approve
Brigham.ai executes the payload that matches the hash.
Built for mail scanners.
Security gateways open every link in every email. So an approval link opens a confirm page and acts only on a click or POST. Scanning it does nothing.
What cannot approve.
Tool output
Nothing returned by a tool can grant approval.
Email and web content
Untrusted content can propose an action, never trigger one.
The agent itself
Sensitive actions always need a human decision.
Questions, answered.
What counts as sensitive?
External email, deletes, money, ad budgets, publishing, production deploys, contracts, refunds and permission changes.
Can I pre-approve?
Reversible writes can be enabled by an admin per category. Sensitive actions cannot.
Keep reading.
Delegate without the dread.
Get started today.