HOL Guard

AI coding agents can run shell commands, read credential files, connect to MCP servers, and install packages on your behalf. HOL Guard is a local-first runtime control layer that intercepts those actions before they run and applies your policy: allow, block, or ask you first. It runs on your own machine alongside the agent, so you keep the decision. The core is open source under Apache-2.0 — install with pipx install hol-guard && hol-guard init. Teams that want a hosted, fully managed option can add paid Guard Cloud. HOL Guard is not a gateway or network firewall, it does not claim to stop every prompt injection, and a scan is not a safety guarantee. Made by HOL.