agent wallet

An agent wallet is where the money an agent spends is held and where the limits on that spending are set. The wallet holds the credential and the agent presents something derived from it. What differs between rails is who approves a spend, and what the agent is given to present.

On the card side the wallet keeps the credential and issues a scoped grant instead of it. Stripe's documentation describes shared payment tokens as scoped grants that let an agent use a customer's payment method, each carrying usage and expiration limits, issued from a wallet built for agents. That wallet's own description is that payment credentials are never exposed, and that each purchase is confirmed by the person in the wallet's app. Finer-grained spending controls are listed as not yet available.

On the stablecoin side the wallet is a signing key. Under x402's exact scheme the agent signs an authorization for a stated amount to a stated address, and the signature is what travels. The funds move when a facilitator submits that authorization. The limit is the authorization itself — an amount, a recipient and a validity window — rather than an approval step at the moment of spending.

The two models put the control in different places, which is worth knowing before either is handed a budget. A wallet that confirms each purchase can decline one. An authorization that has already been signed is a signature somebody else can submit, and what arrives at the endpoint either satisfies the quoted terms or does not.

→ probe402 runs a settle leg of its own, and the network and asset that leg will accept are set out in the method.

Sources

Related agent · settlement · EIP-3009 · buyer

Glossary