Lumière PayCheck overview
AI agents can now pay for APIs on their own, using x402: a request comes back "402 Payment Required" with a price, and the agent pays in USDC and retries. That's powerful, and risky. Is the seller real? Will it actually deliver? Has its payout wallet been swapped? And who decided this agent could spend the money?
Lumière PayCheck answers "should this payment happen?" before the money moves, and keeps a record of every answer your auditors can verify.
How it works
- We watch the market. Every x402 endpoint listed in the x402 Bazaar is checked every 30 minutes: is it up, what does it charge, which wallet gets paid. We also pay endpoints ourselves to confirm they deliver, and read real payments on-chain to see which sellers actually earn.
- Your agent asks before paying. It sends the endpoint, amount, and payout wallet to PayCheck. We check it against our trust data and against your company's rules.
- We answer allow, deny, or review. An "allow" comes with a signed receipt your wallet can require. A "review" waits for a person on your team to approve it.
- Everything is recorded. Each decision is stored with its full inputs, can be replayed exactly, and is chained so any later edit or deletion is detectable.
Who it's for
- Teams deploying AI agents that spend money: give each agent its own scoped key and limits instead of a shared wallet.
- Finance and compliance: company-wide caps, budget alerts, spending by team, read-only access, and an audit trail you can verify.
- Security and IAM teams: roles with separation of duties, IP allowlists, key-less sign-in for agents, a freeze switch, and decision streaming to your security tools.
- Agent developers: a free API and MCP server to check any x402 endpoint before calling it.
What makes it different
- Before the payment, not after. We decide whether money should move; we never hold, route, or escrow it.
- Evidence, not opinions. Grades come from continuous monitoring, our own paid delivery tests, and real on-chain payments, not from sellers' self-descriptions.
- Built like an identity system. Scoped credentials, least privilege, approvals, and an audit trail: the controls companies already apply to people, applied to agents.
Where to go next
Questions? Email hello@lumierepaycheck.org or talk to us about Enterprise.