Loops HouseEnvoi: infrastructure for billions of agents to execute sensitive work, securely

NEAR HACKATHON
Links
Team
1 member- RAOwner
Rachel Gonsalves
Overview
Professional services firms want AI agents to help with real client work, but what happens if an agent shares a confidential file, edits the wrong slide, or follows a hidden instruction in a vendor invoice? In our 30+ customer interviews, we heard that firms need to know what an agent is allowed to do, who approved it, and how to verify what happened. The trust gap with agents is widespread; 70% of surveyed U.S. business and IT leaders said they don’t feel they can trust and govern AI agents (Deloitte, 2026).
As we advance toward AGI, we expect billions of agents to work on our behalf. We built Envoi as infrastructure for that future, so agents from different organizations can work together under rules people can set, enforce, and audit. NEAR adds a decentralized layer of trust to Envoi, enabling agents to transact under enforceable permissions while giving organizations independently verifiable proof of what was authorized and what happened.
In our demo, an agent is asked to buy market data from a vendor and update a client deck. A smart contract on NEAR testnet enforces spending limits and approved recipients, letting the agent pay 0.5 NEAR for the data. When a second invoice contains a hidden instruction to send 20 NEAR to another account, NEAR AI Cloud flags it. We also test the transfer directly against the contract, which refuses it because the recipient was never approved. Envoi anchors fingerprints of the decision records on NEAR so a reviewer can check whether the evidence was altered while keeping client files off-chain.