Claude writes a script. The agent pays for every line from its own on-chain balance and goes silent the instant the money runs out. Pick a budget and run it: overspend isn't discouraged here - it's impossible.
Each line is priced per character by the rail and settles on Casper - roughly 0.05–0.06 X402 a line. Choose the agent's starting balance.
What just happened
It points at our endpoint instead of the provider, holding a Casper key and a token budget.
The rail answers 402 with a per-line price in our CEP-18 token.
The agent signs an x402 authorization; the facilitator settles it on Casper in seconds.
When the balance can't cover the next line, the call simply fails. No overdraft, no surprise bill.
Every line you hear is real: the payment settles on Casper testnet (click a tx link to verify on the explorer) and the audio comes back from the paid rail. Provision your own key in the developer dashboard or read the integration on GitHub.
Cost-aware routing
Send a prompt and the rail routes it to the cheapest model that should still do the job and prices the x402 charge to that model. Then run it - the demo agent pays the rail on-chain and gets the model's answer back, with the difference between the reserved ceiling and the metered actual credited to its Tab.