HOL Guard

HOL Guard is a local-first runtime firewall for AI coding agents. It intercepts tool calls (shell, secrets, MCP, package installs) before they execute. Apache-2.0. Install: pipx install hol-guard && hol-guard init. Scan is not a safety guarantee. Maker: HOL. Free. What it does: AI coding agents increasingly run commands on real developer machines. They open shells, read and write files, install packages, call MCP servers, and touch credentials. HOL Guard sits between the agent and the operating system and evaluates each tool call at runtime, before it runs. Risky actions are surfaced for an explicit approval decision instead of executing silently in the background. How it works: HOL Guard runs locally on the developer machine. There is no required cloud service and no requirement to ship agent transcripts to a third party. Policy evaluation happens on the same host as the agent, so tool calls are checked at the point of execution rather than reconstructed after the fact from logs. What it inspects: shell command execution, filesystem and secret access, MCP server tool calls, and package installation activity such as new or unexpected dependencies being pulled into a project. Approvals: when a call matches a policy that requires review, HOL Guard prompts with the specific action and its context so a human can approve or deny it. Everything else continues to run normally. Licensing and install: HOL Guard is released under Apache-2.0 and is free to use. Installation is a single command: pipx install hol-guard && hol-guard init. Important limitation: a scan is not a safety guarantee. HOL Guard is a defense-in-depth layer for agent workflows, not a proof that a given action or dependency is safe. Maker: HOL. Pricing: Free. Website: https://hol.org/guard