Get runtime security in under a minute.

Download Guard Desktop, choose your AI harness, and connect Cloud later if you want shared history.

Free and open source. Make your AI ask before destructive commands.

Local protection starts in under a minute

Download Guard Desktop

Local by default

One installer. Local protection. Guard Core still makes every security decision, and Guard Cloud stays optional.

Prefer the command line?
set -o pipefail; curl -fsSL https://hol.org/guard/install.sh | bash -s -- --mode install --verify
hol-guard init

Installation FAQ

Download Guard Desktop for Mac, Windows, or Linux from this page, then open the app. The command line remains available as a secondary path. Guard configures your AI agent to route tool calls through the local proxy. The entire setup takes under a minute.

Yes. Download the Apple silicon macOS app, Windows x64 MSI, or Linux x86_64 AppImage from the install page when those releases are published. The command line remains available as a secondary path. Guard Desktop is a local control center; Guard Core still makes the security decisions, and Guard Cloud stays optional.

Guard requires a supported AI agent and a computer. Desktop installers cover Apple silicon macOS 13+, Windows x64, and Linux x86_64. The command line needs Python 3.10+. No cloud account is needed for the free tier.

Guard checks supported shell, file, MCP, skill, prompt-sensitive, and package actions before side effects. It blocks denied actions and asks for your decision when policy requires review.

Yes. Guard can block denied actions and pause risky commands for approval before they run. Your local policy decides which actions need a warning, an approval, or a block.

No. Guard Local enforces policy on your machine without a cloud connection. Guard Cloud is optional for shared history, team policy, dashboards, and connected-device workflows.

The install page has guides for Codex, Claude Code, Cursor, Gemini CLI, OpenCode, Hermes, OpenClaw, GitHub Copilot CLI, Antigravity, Kimi, Grok, Pi / Oh My Pi, Zcode.

No. Guard adds a local review layer in front of supported harness actions. Your harness stays installed, and its native approval flow remains available where that harness provides one.

Guard installs managed launchers or hooks for the harness you choose and keeps local Guard state for decisions. The harness guide on this page lists the files and integration mode for each tool.

Use the update command shown in the Verify and manage section. Choose the macOS and Linux command or the Windows PowerShell command, then run hol-guard doctor to check the result.

Yes. Guard is free and open source. The local runtime can protect your AI harness without a paid plan or cloud account; optional Cloud features add shared workflows.

After installation, trigger any tool call from your AI agent (e.g., a shell command). Guard should display an approval prompt before the action executes. You can also check the Guard status indicator in your terminal.

Run the Guard uninstall command, which removes the proxy and restores the original agent configuration. Your audit log is preserved locally and can be deleted manually.