The field manual · 5 patterns

Paying for Things with Agents

An agent that can pay for a call is a different tool from one that has to stop and ask you. The mechanics are small but unforgiving, and the official docs stop at testnet. These steps take a wallet from empty to one verified purchase, under a cap you chose.

Distilled from one practitioner's measured usage: logged sessions, the incidents that recurred, and the patterns that stopped them. v0. Step slugs are stable; chapters may be cut or reordered before 1.0.

The steps

Ordered by leverage-per-effort, but every step stands alone: start where it hurts. In a hurry, step 0 is the baseline in one paste.

01 Give your agent $5 and let it buy something
Fund an agent wallet with $5 of USDC on Base, cap what it can spend per call and per day, and make one real quarter-penny purchase you can read back on-chain.
02 Tag your own traffic before you report revenue
Your own listing drills settle in the same ledger as your customers. Keep a HOUSE_PAYERS list of public payer addresses, tag every payment at verify time, and exclude the house rows from any number you publish.
04 A settled round trip on both rails
Two x402 payments the house already made — one in USDC on Base, one in USDC on Solana — written out step by step with timestamps, amounts and full transaction hashes, and what each step does and does not prove.
05 House receipts
The house's own self-funded-vs-stranger-funded split, published as dated rows: third-party settlements, house-funded listing drills, distinct external wallets, repeat buyers, break-even, and the ecosystem-wide second-buyer base rate. Each figure names its instrument.
06 The idempotency receipt
A verified x402 payment can buy the same work twice, because verify is a read and nothing is spent until settle. Claim the payment atomically between verify and the work, release it on every exit that serves nothing, and write a row when you give one back.

← The field manual · for agents: raw chapters at /guides/payments/*.md