AgentGuard

AgentGuard is an autonomous AI treasury governance framework built on Arbitrum. It enables AI agents to earn verifiable on-chain credentials through completed work, build reputation over time, and gain limited treasury authority based on demonstrated performance.

When an agent completes a task, its output is evaluated, recorded on-chain as a credential, and contributes to its reputation score. As agents progress through credential levels, they become eligible to participate in treasury operations.

Agents can autonomously determine whether treasury funding is required to implement their recommendations. If funding is needed, the agent generates a treasury request while a deterministic policy engine calculates the maximum allowable amount based on the agent’s credential level, evaluation score, and available treasury funds.

Human governance remains in control at all times. Treasury requests must be reviewed and approved, rejected, or executed through governance controls, ensuring accountability and preventing unrestricted autonomous spending.

By combining AI agents, on-chain reputation, credential-based permissions, and treasury governance, AgentGuard provides a framework for safely integrating autonomous agents into decentralized organizations.

Started with a basic agent credentialing concept and evolved it into a complete autonomous treasury governance system. During the hackathon, I designed and implemented on-chain agent credentials, reputation-based permissioning, credential progression, treasury request workflows, AI-powered treasury decision making, deterministic spending policies, and governance controls for approving, rejecting, and executing requests.

The final system enables AI agents to earn on-chain credentials through completed work, build reputation over time, and autonomously create treasury spend requests that remain subject to human governance oversight.

Currently focused on product development and validation. Not actively fundraising at this stage.