When you let Claude Code, Codex, Cursor or a similar AI coding agent work in your repo, it can run commands, open files and pull in dependencies on its own. HOL Guard is a local-first runtime control that sits between the agent and those risky actions. It looks at shell commands, reads of secrets and credential files, MCP tool calls and package installs before they execute, and when an action needs a human decision it stops and asks you to approve or block it. Decisions are recorded locally so you can review what happened later. The core is an Apache-2.0 open-source Python CLI; teams that want more can add the optional paid Guard Cloud, so pricing is freemium. Scope, stated plainly: HOL Guard is not a gateway or a network firewall, it does not fully prevent prompt injection, and a scan result is not a safety guarantee. Get started: pipx install hol-guard && hol-guard init
Pricing: Freemium (Apache-2.0 open-source core; optional paid Guard Cloud).
Website: hol.org/guard
Reactions & commentsFeedback
Share your thoughts, bug reports, or suggestions directly with the developer
Log in to share your feedback
Log in to leave feedback