AI · agent accounts
A walled account
for every agent
Open a sub-account for an agent and give it a mandate: what it may spend, on what, up to what limit, for how long. The limit is enforced in settlement logic, not by trusting the model to behave.
01
Issue the account
A human opens a walled sub-account funded to a ceiling. The agent transacts freely inside that box and cannot reach anything outside it.
02
Write the mandate
Categories, a per-period ceiling and an expiry, signed as a cryptographic mandate in the AP2 style — not a prompt the model is asked to respect.
03
Enforce below the model
The check sits under the AI, in settlement. A jailbroken or confused agent still cannot exceed its box. Building that enforcement correctly is the work that matters.
A procurement agent's mandate
Ceiling50,000 per month, hard
ScopeApproved vendor categories only
Expiry90 days, then it stops
StandardAP2-style signed mandate
EnforcementSettlement logic, below the model