HOL Guard

HOL Guard is an open source security layer that monitors what AI coding agents attempt to do on your computer. It can inspect shell commands, sensitive file reads, package installations, MCP configuration changes, plugins and skills before the action runs. Each request can be allowed, blocked or sent to you for approval, with local audit receipts recording the decision.

The software works with Codex, Claude Code, Cursor, Gemini CLI and several other agent environments. Integration depth varies because each agent exposes different hooks. Some receive detailed checks before tool use, while others depend more heavily on wrappers and approval prompts.

Guard Local runs offline without an account. Policies, receipts and configuration stay on your machine, making it suitable for developers and small teams that want local control over agent permissions. An optional hosted service adds shared policies, central receipts and team administration. That cloud component is not available for self hosting.

HOL Guard is a policy layer rather than a sandbox or network firewall. It cannot inspect actions an agent does not expose through supported hooks, and it does not provide complete protection against prompt injection. The project is also developing quickly, so cautious users should pin versions and verify release attestations before updating. The pricing model is freemium, with the local runtime available free and collaboration features provided through paid cloud plans.

Disclosure: This post contains affiliate links. If you purchase through my link I may receive a commission at no extra cost to you.