Mandate is a policy firewall for AI-agent finance. It keeps user funds in a smart account and lets agents act only when onchain rules approve, blocking unsafe actions before funds move.
Mandate is a policy firewall for AI-agent finance. It lets a user fund a MandateAccount, while an AI/demo agent can only act through a constrained session key. Every proposed financial action is checked onchain before execution against asset, adapter, trade-size, turnover, exposure, nonce, deadline, and signed price-row constraints.
The demo shows two paths on Robinhood Chain testnet: a small TSLA buy that is approved and executed after policy preview returns OK, and a dangerous 500 USDG → TSLA action that is blocked because it exceeds the exposure boundary. The agent can propose, Mandate decides, and transaction receipts prove the outcome.
<p> During the Buildathon, I built the full Mandate MVP: smart contracts, deployment on Robinhood Chain testnet, a public landing page, and a live demo console.</p><p>Implemented features include:</p><ul><li><p>MandateAccount policy engine</p></li><li><p>MockERC20 demo assets: USDG, TSLA, AMD</p></li><li><p>MockAMM and approved swap adapter</p></li><li><p>signed demo price feed</p></li><li><p>server-side price signer API</p></li><li><p>server-side demo-agent relayer using a constrained session key</p></li><li><p>wallet-based mint / approve / deposit flow</p></li><li><p>policy preview before execution</p></li><li><p>approved safe action execution</p></li><li><p>blocked dangerous action proof</p></li><li><p>activity/evidence UI with transaction links</p></li></ul><p> The live demo successfully demonstrates that user funds remain in MandateAccount, the agent does not hold the owner key, approved actions can execute, and out-of-policy actions are blocked before funds move.</p>
<p>Mandate is currently unfunded and independently built. No institutional or external funding has been raised.</p>
