Review what your AI coding agent wants to run before it runs: local allow/deny for shell, secrets, MCP tools and installs.
This project is scheduled for launch
Launch date: Tuesday, December 29, 2026 at 08:00 AM UTC

HOL Guard is local-first runtime control for AI coding agents such as Claude Code, Codex and Cursor. Before the agent runs a shell command, reads a secret or credential file, calls an MCP tool or server, or installs a package, Guard checks the action on your machine and, when it needs a decision, asks you to allow or deny it. Each decision is kept in a local record so you can review what the agent tried and what you approved.
It is a Python CLI for macOS and Linux: pipx install hol-guard && hol-guard init. The core is open source under Apache-2.0, with an optional paid Guard Cloud (freemium).
Scope: HOL Guard is not a gateway or a network firewall, it is not a complete defense against prompt injection, and a scan result is not a guarantee of safety.
Comments will be available once the project is launched.
Visit Website
hol.org
Michael Kantor
@holguard
Share this launch.