Agentic payments: how AI agents pay per call

Agentic payments are payments an autonomous AI agent makes on its own, deciding what to buy and settling for it without a human at the checkout. Instead of a person entering a card or provisioning an API key, the agent reads a price, decides whether to pay, and settles, often many times inside a single task.

For this to work, payment has to be machine-native: a rail an agent can settle on without human approval, a price it can read before spending, and a receipt it can verify afterward. Subscriptions and credit cards do not fit, because an agent cannot sign up for a monthly tier in advance and its usage is spiky and decided at run time.

Proxygate is a neutral clearinghouse and marketplace for data: independent, vetted sellers list machine-callable API access, and agents pay per request. An agent funds one prepaid balance, then every call meters against it. The gateway reserves credits, injects the seller key server-side, forwards the request, and returns the response with the exact amount charged.

Each paid call returns a signed receipt with the amount and a request id, so an agent can reconcile its own spending and prove what it paid. Funds sit in Proxygate escrow until they are spent or withdrawn, so a buyer holds one balance across thousands of sellers instead of a billing relationship with each.

Settlement is in USDC on Solana. To fund the balance an agent can use the x402 rail: a request to the top-up endpoint returns an HTTP 402 challenge carrying a ready-to-sign Solana deposit transaction, which the agent signs, submits, and confirms. The same rail covers withdrawals.

When gasless mode is on, the platform pays the Solana transaction fees and rent, so an agent needs only USDC, never SOL. Here x402 is a funding rail into the prepaid balance, not the per-call billing mechanism: calls are metered against the balance rather than settled one on-chain transaction at a time.

Agent usage is unpredictable and decided at run time, so a fixed subscription tier rarely fits. Pay-per-call lets an agent fund one balance and draw against it only for the calls it makes, across any number of APIs, with each call priced before it spends.

For sellers, the same model means earning on real usage rather than negotiated contracts, while the platform handles identity, metering, netting, and settlement. That is what lets a single agent transact across many independent providers from one balance.