How to monitor an AI agent's wallet using only its public address

What a public address gives you

Every USDC transfer in or out of an address is recorded on-chain with a sender, a recipient, an amount, a block or slot time and a transaction hash. That is the whole payment history of the wallet, readable by anyone.

On Base this means ERC-20 Transfer events for the USDC contract. On Solana it means SPL token transfers touching the wallet's USDC token account, which is derived from the owner address and the USDC mint.

What it does not give you

The chain does not say why a payment was made, which task triggered it, or which of several agents sharing a wallet sent it. Reason and task have to come from your own logs. Attribution to an agent only works cleanly when each agent has its own address.

Why read-only is the right default

A monitoring tool that holds no keys cannot move funds, so the worst case from a compromised monitor is a wrong chart, not a drained wallet. Reading a public address also means setup is one paste, with no permissions to grant and none to revoke later.

A practical setup

Give each agent a dedicated wallet, record the address next to the agent's name, and point a read-only tracker at those addresses. Set a ceiling per agent, and alert when spend approaches it. AgentCap does exactly this for USDC on Base and Solana: paste an address, get a ledger in about a minute.