The problem Wallexa solves

initial commits: FE, contracts, agent-showcase

Wallexa makes AI agents safer to use onchain by turning a regular hardware-wallet into a modular smart wallet with scoped agent access.

Today, giving an AI agent blockchain access usually means handing it a private key with full control, or creating a separate wallet that must be constantly funded and often lacks the permissions, reputation, or app history of the main wallet. Wallexa solves this by keeping the hardware wallet as the root authority while using EIP-7702, ERC-4337, and ERC-7579 modular account infrastructure to issue limited agent keys.

Agents can transact from the user’s real wallet address, but only within boundaries enforced onchain: token spend limits, ETH budgets, contract allowlists, function-level permissions, time windows, and revocation. Example roles include stablecoin payment assistants, DeFi rebalancers, rewards claimers, treasury payers, gas runners, governance voters, and many more.

Users no longer need to expose their main private key or constantly refill separate agent wallets. The wallet owner can sleep calmly knowing their agents can keep working from the secure hardware-wallet, but cannot bypass the rules set for them.

Challenges we ran into

One of the biggest challenges was that the idea started from the hardware itself. The Idea was born at the hackathon when we received a small hardware wallet with open-source firmware, but the firmware did not support signing the new EIP-7702 authorization needed to delegate an EOA and turn it into a smart account. To make Wallexa possible, we had to fork the wallet firmware and add custom support for signing EIP-7702 authorization lists directly from the hardware wallet.

Another major hurdle was the permission system. We could not find an existing ERC-7579 validator module that matched the exact access-control model we needed for AI agents: token spending limits, native ETH budgets, contract allowlists, function-level restrictions, time windows, and revocation. So we built and deployed our own Solidity ERC-7579 validator module that enforces these boundaries onchain.

The final challenge was proving that the system actually works with real agents. We built a small agent demo repo where agents can be initialized with the generated access keys and attempt transactions from the user’s wallet address. This let us demonstrate both sides: transactions inside the agent’s permissions succeed, while transactions outside its boundaries fail, even if the agent is instructed to perform them. Most importantly, the successful transactions are still made from the hardware wallet’s address, showing that the agent can operate through the user’s real wallet without ever receiving full control over it.