AI agents are becoming more autonomous.
They can browse, trade, call APIs, execute tasks and interact with other services.
But one problem remains:
How do you give an AI access to money without giving it unrestricted access to your wallet?
We built Wallet Agent AI to explore that problem.
Instead of giving an AI access to your main wallet and hoping it behaves, the agent gets its own wallet with programmable permissions.
For example: Spending limits to control how much the agent can spend.
Transaction limits to restrict the maximum amount per transaction.
Whitelists to limit which destinations and assets the agent can use.
Freeze and revoke controls to stop the agent immediately.
Owner-controlled permissions so the agent never gets unrestricted control of your funds.
On-chain authorization enforced on L1 Radix smart contracts, rather than relying only on instructions in an AI prompt.
DeFi capabilities including swaps, liquidity, staking and other Radix applications.
AI integrations designed for autonomous agents, not just human users.
The interesting part is not simply giving an AI its own wallet.
It is giving the AI limited financial capabilities.For example:
You have 100 XRD or HUSDC or whatever token.
You can spend a maximum of 5 XRD per transaction and interact only with these approved applications.
If the agent then tries to spend 20 XRD, it can make the request, but the transaction is rejected. The AI can try.
The blockchain does not have to trust it.
This is not yet an AI that can shop on Amazon or make payments across the internet.
The goal is to build the infrastructure for programmable financial agents first, then connect them to real services and agent-to-agent economies.
The project is open source because we want other developers to experiment with the idea and build on top of it.
Wallet Agent AI: https://wallet-agent-ai.github.io/ The GitHub repository and SDK are linked from the project.
What would you trust an AI agent to do with money if you could define and enforce its permissions on-chain?