Autonomous agents are starting to pay for things on their own — APIs, data, services. PayPerimeter sits between the agent and its wallet. You set the rules; the blockchain enforces the ceiling. Even if our software is bypassed entirely, the loss is bounded.
Try it live → set a cap and turn the agent loose.

Set a daily cap, then let the agent run.
Software fails in ways people don't. It doesn't get tired, hesitate, or notice that something feels wrong — it just repeats, at machine speed, until something stops it.
A human makes a mistaken purchase once. A looping agent makes the same mistaken purchase hundreds of times a minute, and nothing in the payment rails objects.
An agent that reads the open web can be talked into things by the content it reads. If that agent also holds a key, persuasion becomes spending.
A limit written in the same program the agent runs is a limit the agent can route around. Real enforcement has to live somewhere the agent can't reach.
Your funds sit in a vault the agent never touches. PayPerimeter releases small amounts to the agent's spending key — and the maximum it can ever release is fixed on-chain, by an audited contract, not by us.
Before anything is signed, PayPerimeter checks the payment against the rules you wrote.
The vault grants PayPerimeter permission to move at most a fixed amount per period. That rule lives in an audited smart contract.
The point isn't that we promise to stop overspending. It's that past a certain number, there is nothing left to promise — the payment simply cannot happen.
This isn't a concept. The full system runs end to end on Base Sepolia today — real payments, real on-chain caps, real refusals when an agent tries to go past them.
Stated plainly: PayPerimeter is early. Everything above runs on a test network, not with real money, and the code has not yet been through an external security audit. We'd rather tell you that up front than have you find out later — a company built on enforcing limits should be honest about its own.
Agents paying for things is going from a demo to a default. When that happens, the question stops being should this be governed and becomes what is actually enforcing it.
The policy engine, the signed audit log, and on-chain spending caps — working together, proven on testnet.
An external audit, then live networks. Alongside it: the dashboard a team needs to set policy, review approvals, and watch what their agents are actually spending.
Every autonomous agent trusted with money needs something enforcing its limits and recording what it did — the way a compliance system sits under every employee with a company card. We want to be that layer.