A smart account an AI agent can operate but cannot drain. The human owns the account; the agent's key can only make calls the on-chain policy admits (target, function and payee allowlists, rolling spend windows, rate limits, approval ceiling, expiry, guardian breaker). simulate() returns the same refusal execute() would revert with. Verified on Base Sepolia, Whitechain Sepolia and GOAT testnet; a prompt-injected agent tried six ways to drain it and lost nothing. Ships with an MCP server, a Claude Code plugin, a GOAT SDK wallet provider, and rein scan, which reads an agent wallet's history on Base and prints the policy that history supports.

Fund this project

Unverified URL

The funding manifest has not provided proof via wellKnown that this link is associated with it. Learn more.

Continue