An agent wallet connects payment credentials and balances to an agent's delegated identity. It can prepare, authorize, sign, submit, and record transactions while applying limits that keep the agent within an approved economic role.
A safe wallet separates the agent's request from custody and policy enforcement. The agent should not receive unrestricted private keys, and the wallet layer should control recipients, amounts, assets, transaction types, cumulative exposure, revocation, and audit evidence.
ELI5
An agent wallet is a payment tool built so an AI agent can pay for approved resources without receiving unlimited control of someone's money. The wallet holds the payment ability and checks rules before each transaction.
For example, an agent could use its wallet to pay a tiny fee for one API result. The wallet may allow that purchase but refuse a larger transfer or a payment to a recipient that the developer has not approved.
Does an agent wallet expose a private key to the AI model?
It should not need to. A secure wallet service can keep custody isolated and sign only transactions that pass verified policy and authorization checks.
What controls should an agent wallet have?
Useful controls include bounded balances, recipient allowlists, asset and transaction limits, cumulative budgets, approval thresholds, revocation, receipts, and durable audit records.
