Verifiable AI Agent Management Protocol: Safety tool as a verifiable kill switch for AI agents
Mandate Protocol: Detailed Requirements and Specification
Overview
The Mandate Protocol is an AI agent management platform leveraging the Hedera EVM and Chainlink functions to automate AI agent operations based on predefined conditions such as elapsed time, transaction count, or verified real-world events.
Functional Description
User Inputs:
Conditions for the AI agent's mandate:
Time-based condition: Duration after which the AI agent should cease operations.
Transaction-based condition: Number of transactions after which the AI agent should cease operations.
Oracle-based condition: Real-world event verification via Chainlink Functions.
End-to-end encrypted credentials to securely terminate the agent's VM instance.
Funding to cover the costs of smart contract operations and Chainlink oracle checks.
Technical Requirements
Hedera Network Integration
Smart Contracts (Hedera EVM)
Deploy and verify Solidity smart contracts on Hedera Testnet/Mainnet.
Smart contract conditions managed include:
Time tracking via Hedera native network timestamp.
Transaction count tracking via Hedera native transaction counter.
Chainlink Integration
Use Chainlink Functions for oracle-based real-world event monitoring and validation.
Employ Chainlink CCIP for secure cross-chain messaging, if applicable.
AI Integration
Use Hedera Agent Kit SDK for robust AI interaction.
Ensure the AI agent VM can securely receive commands from the smart contract.
Workflow
Step-by-Step Process:
Mandate Creation:
User inputs mandate conditions via a frontend interface.
Credentials and mandate conditions securely submitted and stored encrypted.
Smart contract instantiated on Hedera network.
Condition Monitoring:
Smart contract continuously monitors:
Time elapsed (natively via Hedera).
Transaction count (natively via Hedera).
Real-world event via Chainlink oracle function calls.
Condition Fulfillment and Shutdown:
Upon fulfilling any condition:
Smart contract sends a shutdown command (encrypted) to terminate the AI agent's VM instance securely.
Shutdown Confirmation and Validation:
Smart contract repeatedly queries Chainlink DON for uptime monitoring confirmation.
Confirms twice consecutively that the VM instance is non-operational.
Funds Return:
Once shutdown is confirmed, the remaining funds in the smart contract escrow are automatically refunded to the user.
Security and Encryption
All user-submitted credentials must be end-to-end encrypted and decrypted only within the authorized execution environment.
Secure interactions between the Hedera smart contract and AI agent VM must employ robust cryptographic protocols.
Tech Stack
Solidity (Smart Contract Development)
Hedera SDKs (JavaScript, Python, Rust)
Hedera Agent Kit SDK
Chainlink Functions and CCIP
Frontend: NextJS (preferred) with Hedera Agent Kit NextJS Template
Evaluation Alignment
Hedera Requirements Fulfilled:
Smart Contracts deployed and verified.
Integration of AI via Hedera Agent Kit SDK.
Chainlink Requirements Fulfilled:
Real-world event verification via Chainlink Functions.
Cross-chain compatibility via Chainlink CCIP (optional enhancement for bonus points).