HOL Guard
An antivirus-style watchdog for AI coding agents that runs on your own machine, checking every shell command, file edit and package install before it happens and blocking anything that looks dangerous.
🔗 Visit HOL Guard
Description
AI coding assistants like Claude Code, Cursor or GitHub Copilot CLI can now run shell commands, edit files and install packages on their own — which is powerful, but also means a bad suggestion, a prompt injection, or a hallucinated command can do real damage to your machine before you notice. HOL Guard installs like antivirus software for these agents: it sits locally on your computer and inspects every shell, file, MCP, skill and package action the agent tries to take, blocking the risky ones and asking for your approval on anything flagged.
It's open source under the Apache 2.0 license, works locally by default with no cloud account required (cloud connectivity is optional), and supports over 13 AI agent platforms with desktop installers for macOS, Windows and Linux. Beyond blocking obviously unsafe commands, it detects secret exposure (accidentally printing an API key), flags prompt injection attempts, and scans installed packages and MCP servers for known-malicious supply-chain risks. Everything it catches is written to a local audit log that survives even if you uninstall the tool, so you keep a record of what your agents actually tried to do.
💬 Our review
The short version: HOL Guard is a free, local-first "antivirus for AI agents" — it doesn't ask you to trust a cloud service with your commands, it just watches what your coding agent does on your own machine and steps in before something dangerous executes.
The local-first design is the main selling point over a cloud-based alternative: your shell commands, file paths and secrets never have to leave your machine to be checked, which matters if you're working with client code or sensitive infrastructure. Being open source (Apache 2.0, published on PyPI as hol-guard, 2,300+ commits and 460+ GitHub stars) with support for 13+ agent platforms and desktop installers for all three major OSes makes it easy to actually adopt rather than just read about. The homepage itself returned an HTTP 403 to automated fetching during this review, likely bot protection rather than a sign of anything wrong, but it's a reminder to verify the GitHub repo (hashgraph-online/hol-guard) directly if you want to confirm activity before installing. Its closest comparison in this catalog is Phinq, which takes a more proxy/SDK-based approach aimed at agent pipelines in production rather than a desktop app for individual developers — pick based on whether you want to protect your own coding sessions (HOL Guard) or gate a deployed agent's actions (Phinq).
💰 Pricing
📊 Global score
🤖 AI-enriched data
Gratuit et open source sous licence Apache 2.0, fonctionne localement sans compte cloud requis. Connectivité cloud optionnelle. Aucun palier payant publié.
Pros
Gratuit, open source, licence Apache 2.0
Fonctionne localement, aucun compte cloud requis
Compatible avec 13+ plateformes d'agents IA, installeurs macOS/Windows/Linux
Détection d'exposition de secrets, d'injection de prompt et de paquets malveillants
Cons
Page d'accueil bloquée par protection anti-bot lors du test automatisé (à vérifier manuellement)
Aucun palier payant ou support commercial publié pour les équipes qui en auraient besoin
Chevauchement avec Phinq, alternative open source proche mais orientée production/pipeline