Mandate Agents
Agent executes real on-chain financial actions for you, inside limits you set, with a ledger
Demo
About
Mandate is bounded delegation for AI agents. You control what contracts an agent can call, how much USDC it can move, and which actions need your Ledger approval. The smart account enforces these limits on-chain, not the AI. Tell it: “Get me 100 USDC on Arc by Friday to pay Acme. Don’t sell my ETH.” It checks your positions, compares borrow rates, creates and simulates a plan, then executes safe steps on its own. For irreversible actions like bridging via CCTP, it pauses for a plain-text Ledger approval. The contract verifies the exact approval on-chain before executing. Everything is enforced on-chain: spending caps, allow-lists, replay protection, and guardian approvals. The agent never sees your keys or writes calldata.
How it's made
Built with Solidity + Foundry + OpenZeppelin, with a TypeScript/viem SDK powering deterministic action adapters and agent tooling. The core idea is bounded, verifiable execution: the AI only creates an intent; deterministic code builds the actual calls, while the smart account enforces allow-lists, per-tx/daily caps, replay protection, and guardian approvals. We use Circle CCTP v2 for Base Sepolia → Arc transfers and The Graph for standardized lending-rate data across protocols. Safe steps execute automatically; irreversible steps trigger a plain Ledger approval that the contract reconstructs and verifies byte-for-byte on-chain.