Concepts and protocols
Custodial vs non-custodial wallets for AI agents
Who holds the money and the keys when an AI agent pays: custodial balances, agent-held keys, and owner wallets with delegated limits compared.
In a custodial setup a provider holds the funds and the agent spends from that balance; in a non-custodial one the money stays in a wallet you control and the agent either holds a key itself or requests payments that you, or a capped delegation, authorise.
Why is this harder than it sounds?
The question decides what you are trusting. With custody, you trust the provider to hold funds safely and apply your rules, and you usually top up a balance in advance. With an agent-held key, you trust the agent and everything it reads. With an owner wallet and delegated limits, you trust a narrower thing: a cap enforced on-chain plus whoever decides destinations inside it. None is risk-free; they move the risk to different places, and the right answer depends on amounts, frequency and how much you want to approve yourself.
How do you do it?
- Custodial balance. You deposit with a provider; the agent spends from that balance under the provider's rules. Convenient, but your money sits with someone else.
- Agent-held key. The agent controls its own address. Simple, but the key is full control and can leak through the agent's context.
- Owner wallet, owner signs. The agent requests; you sign each transfer in your own wallet. Safest, but every payment needs you.
- Owner wallet, capped delegation. You grant a limited allowance or smart-contract permission so small payments can go through unattended, and sign larger ones yourself.
- Ask the same questions of each. Who can move the money? Up to how much? To whom? Who can change those answers, and is it logged?
What it does not mean
- Non-custodial doesn't mean no trust. A delegated spender is trusted within its cap.
- Custodial isn't automatically unsafe. It means you rely on the custodian's controls and solvency.
- A smart wallet isn't non-custodial by definition. Check who holds the keys that control it.
Common questions
Which wallets work with Veyra?
MetaMask, Base / Coinbase (passkey) and WalletConnect for on-chain transfers in USDC on Base. Phantom can be linked for identity only.
Is this financial advice?
No. It explains how custody arrangements differ. Which is right for you is your decision.
Where does Veyra fit?
Veyra is non-custodial: at no point does a payment pass through a Veyra-controlled balance, and it never asks for a seed phrase. It uses the last two models. Above your auto-approve limit, and for every payment on Free, you sign the transfer in your own wallet and Veyra verifies the on-chain result. On Pro and Team, small payments can run from a capped USDC allowance to Veyra's relayer, which pays from your wallet straight to the recipient; Veyra's docs spell out that within the cap you're trusting its policy and relayer key, so auto-pay requires a recipient allowlist.
Sources
Terms on this page
- Token allowance (ERC-20 approve)A token allowance is the amount of an ERC-20 token that an owner has authorised a specific spender address to transfer from the owner's account, set with approve and spent with transferFrom.
- Stablecoin (USDC)A stablecoin is a blockchain token designed to keep a stable value relative to a reference asset, usually a currency such as the US dollar; USDC is a dollar stablecoin issued by Circle.
This page explains how agent payments and the related technology work. It is not financial, investment, legal or tax advice. Protocols and provider features change; check the linked official sources before relying on a detail.