Loops HousePrivatePay Agent

NEAR HACKATHON (Test)
Links
Team
1 member- SAOwner
samimi
Overview
PrivatePay Agent is a restricted and secure payment system for AI agents that allows them to process invoices and make payments without having direct access to a wallet or token balance. The problem this project addresses is how to balance payment automation with the security of assets and financial information. Giving an AI agent payment authority can be risky, as it may extract incorrect information or make an unintended payment under the influence of a malicious request. To address this issue, PrivatePay Agent combines NEAR AI Cloud Confidential TEE with NEAR smart contracts. First, sensitive invoice information such as names, email addresses, account numbers, and IBANs is removed locally. Only the sanitized text is sent to the NEAR AI Cloud confidential environment for processing. The Attestation report and the model's response signature are also verified before the process continues. During the payment stage, the agent only has access to the try_pay method. Before transferring any tokens, the smart contract checks the agent identity, the daily payment limit, the list of approved recipients, and whether the invoice has already been processed. If any of these restrictions are violated, the payment is blocked. The contract itself is responsible for holding the tokens and executing transfers, while the agent never holds any token balance. The project core is developed as a CLI to provide greater security and more precise control over operations and access permissions. We have also designed a Web UI dashboard for monitoring transaction statuses, successful payments, and blocked requests. (just for monitoring) The project has currently been implemented and tested on NEAR Testnet. It supports real payment flows using test tUSDC, onchain restrictions, and structural Attestation verification. The goal of is to provide an environment where AI agents can perform financial tasks autonomously, while control over assets and payment rules remains in the hands of the owner.