HOL Guard is a security tool designed to monitor and control actions that AI agents attempt to perform on a user's computer. Its primary function is to intercept risky commands before they execute, such as destructive file operations, access to saved credentials, software installations, or data transfers. On supported integrations, the user can choose to allow, ask, or block each action, and the tool keeps a record of all activity for later review. The application runs locally on the user's machine, meaning files and data do not leave the computer. The tool offers runtime protection, policy routing, review, and evidence collection. It supports a range of AI tools and extensions, applying policies to each covered action. Users can install Guard directly or first run a risk check to evaluate its behavior in a given environment. A demo mode allows users to explore the interface in read-only fashion before full deployment. Typical use cases include stopping an agent from deleting a documents folder, reading cloud credentials, sending environment variables to an external server, or pushing code directly to a main branch. Because decisions and actions are recorded, users can audit what happened after an interception. HOL Guard is free to start and is available through the HOL website, with options for community involvement and organizational support.