Payments for AI agents · Stellar · USDC
Your agent can spend money.
It can’t spend it without you.
Paydlux puts every agent purchase behind a spending mandate written on the Stellar ledger. The agent asks. The seller waits. You tap Approve on your phone. A hijacked agent can’t overpay, because the cap isn’t in the agent’s software — it’s on the chain.
What actually happens
Five steps. One of them is you.
The agent never holds your money. It only ever asks for permission, and the ask carries the exact request it wants to make — so the thing you approve is the thing that happens.
Agent asks
It calls an API and gets back 402 Payment Required with a price. Same standard every agent payment rail already speaks.
We check the mandate
Is this seller allowed? Is the amount inside the per-call cap? Is there budget left today? Fail any of those and it stops here.
Funds get held
The money moves into a vault that needs two signatures to open. The seller can see it’s there. Nobody can spend it yet.
You approve
A card on your phone: who, what, how much, for which exact request. Approve or deny. Your key never touches the agent’s machine.
Seller paid
USDC lands in the seller’s wallet, the agent gets its data, and a signed receipt goes into the audit log.
The mandate
The spending cap is not a suggestion.
Every other agent payment setup puts the limit in the agent’s code — which is exactly the part an attacker, a bad prompt, or a confused model can reach. Ours lives on the ledger, where the agent has no write access.
// what you set once, in the portal mandate { per_call_max: $2.00 daily_budget: $20.00 total_budget: $200.00 allowed_sellers: [datafeed.example, reports.io] approver: your phone expires: 2026-12-31 } // what the agent can do about that // nothing. it signs requests, not releases.
- Request-bound signaturesThe agent’s signature commits to the exact method, URL and body it is buying. A signature can’t be replayed against a different, more expensive call.
- Two keys, two partiesThe vault is a Stellar account that requires both the agent’s key and yours. We demonstrated a release attempt without your signature being rejected by the ledger, and one signed by the wrong approver being rejected too.
- Denials cost nothingDeny and the money never leaves. It stays in your vault, and the denial is logged with the rest.
- Proof of deliveryOptional escrow: the seller proves what they sent before the release timer lets the money go.
Two sides, one rail
Who plugs in, and how much work it is.
You run agents
One command, run by the agent itself. It creates its own key, registers a mandate with us, and hands you a link. You open the link and approve. Nothing to install on your machine.
- Set limits once, in a browser — no crypto knowledge needed
- Approve from your phone, in the browser, on the web
- Funds stay in a vault only you can release
- Every purchase leaves a signed receipt you can hand to finance
You sell to agents
If you already answer 402 Payment Required, you change one line of config: point the payment check at our vault instead of verifying signatures yourself.
- Settled in USDC, on Stellar, for a fraction of a cent in chain fees
- You see the funds held before you serve the response
- 0.25% seller fee, free while we onboard
- No chargebacks, no card rails, no monthly minimums
Why Stellar
Micro-payments only work if the chain is nearly free.
We tested this on three chains before choosing. A $0.10 API call has to survive the network fee, and dollars have to be real dollars — not a wrapped IOU on a bridge.
Chain fee and USDC circulation verified on public Stellar data, Oct 2026. We also tested XRPL: its native escrow refuses issued tokens, so approval-gated dollar payments aren’t expressible there. That is why we chose Stellar.
Where we actually are
We’d rather tell you now than later.
The full loop runs: agent asks, funds held, human approves from a real wallet, money moves — and a denial stops it. Spending caps enforced. Approval-gated token escrow demonstrated on Stellar testnet. Code is public.
Stellar version of the approval desk on a public testnet, the one-command agent install, and the seller-side config. Early access is for people who want to try it before it’s polished.
A paid independent security audit, and a public mainnet launch. We will not take customer funds before the audit is done and published. If anyone tells you otherwise, they’re not us.
Early access
Tell us what your agent needs to buy.
We’re picking a small number of pilots — teams running agents that pay for data, and sellers who want agent traffic. Send one sentence about your case and we’ll set you up.
Opens your mail app. One sentence about your use case is enough. No list, no tracking, no newsletter.