Agents that pay for their own API calls

Most APIs assume a human signed up, agreed to terms and put a card on file. An autonomous agent has none of that. It has a wallet and a task.

The problem with keys

Give an agent an API key and you have given it unlimited spend on your account until you notice. Most providers' answer is "watch your dashboard", which is not an answer for something running unattended at 3am.

Payment as authentication

GlianaAI answers with HTTP 402 Payment Required. The challenge states the price and the accepted rails. The client pays, retries, and the call goes through. There is no account, so there is nothing to provision and nothing to revoke.

For an agent this is the whole difference: it can discover a capability, find out what it costs, decide, and pay — without a human in the loop and without standing authorisation sitting in an environment variable.

Discovery is machine-readable

The catalog, prices and input schemas are live endpoints, not a PDF. An agent reads what exists and what it costs at runtime, so a new model or a price change needs no client update.

We are indexed by the directories agents actually crawl, and there are MCP servers if your tooling speaks that.

If you do use a key, cap it

Prepaid balances and keys exist for people who want one bill instead of one settlement per call. Every key carries a daily spend cap by default — a leaked key costs a bounded amount rather than the balance.

Failure is free

Requests are validated before payment is taken, and a failed call is refunded. An agent retrying a broken request does not quietly drain a budget.

That is the property that matters when nobody is watching: the cost of getting it wrong is bounded by design, not by vigilance.