HOL Guard

Local-first runtime control that intercepts risky AI agent actions before they execute.

Overview

HOL Guard adds local policy checks to supported AI coding-agent actions. It can allow, ask for approval, or block covered actions such as destructive commands, secret access, software installs and data movement before they run. Coverage depends on the agent integration and action type; it is not a guarantee that every harmful action will be stopped.

Guard Local is free and open source and can operate without a cloud account or internet connection. Supported integrations listed on the official site include Claude Code, Codex, Cursor and OpenCode. Local decisions and activity history help users review agent behaviour.

Optional paid Guard Cloud plans add cross-device history, alerts and team policy features. The pricing page says Cloud sync stores redacted decision summaries rather than source code or secret material.HOL Guard is a desktop security tool that sits between AI coding agents and your machine, intercepting risky actions before they run. It works locally on your computer—your files never leave your machine—and can allow, ask for approval, or block actions like destructive shell commands, secret access, software installs, and data exfiltration.

The tool supports major coding agents including Claude Code, Cursor, Codex, OpenCode, and others. It applies configurable policies to supported action boundaries, maintains an audit trail of decisions, and can optionally sync approvals and policy across devices via Guard Cloud. Guard is free and open source; paid plans add team features and cross-device synchronization.

Key features

Pre-action interception of shell commands, MCP calls, and package changesP
Local-first policy enforcement with optional cloud sync
Audit trail and decision history with evidence exportL
Support for Claude Code, Cursor, Codex, OpenCode, and other agents
Team policy sharing and approval workflows via Guard Cloud
GitHub secret monitoring and credential scanningO
Plugin and MCP supply-chain inspection via plugin-scannerA

Pros

Runs entirely on your machine with no required cloud connection
Free tier includes full local protection and policy enforcement
Open source with transparent security research and benchmarksO
Supports multiple AI coding agents with harness-specific coverage
Minimal performance overhead (millisecond-level latency)L
Detailed evidence and redacted decision history for complianceO

Cons

Cannot prevent prompt injection attacks, only reduce impact at action boundariesA
Coverage varies by agent and supported action types
Requires manual policy configuration for custom workflows
Cloud features (sync, team policy) require paid subscription
Not a replacement for endpoint security, DLP, or secrets management

Use this if

You need pre-action control over AI agent commands and want to keep decisions local. Use if you're concerned about destructive commands, secret access, or data exfiltration from coding agents.

Skip this if

You need complete prompt-injection prevention or a cloud-only solution. Skip if you require network-level controls or don't use AI coding agents.

Best for

Developers using AI coding agents who want pre-action control over risky commandsTeams needing shared security policies for AI agent workflowsOrganizations requiring audit trails and approval workflows for agent actionsUsers concerned about secret exfiltration and destructive commands

Alternatives

Native agent controls (Claude Code, Cursor built-in settings)

Compare HOL Guard alternatives

At a glance

Free plan: Yes
Paid plans: Yes
Setup time: 60 secondsVaries by platform and chosen agent integration
Learning curve: Easy

Get started

curl -fsSL https://hol.org/guard/install.sh | bashDownload Guard Desktop for your platform from https://hol.org/guard/install

Free local protection with no cloud account required. Optional Cloud monthly plans shown on the official pricing page: Solo $4.99/month, Pro $15/month, Team $30/seat/month. Enterprise scope and commercial terms require a separate review. Cloud trial and payment terms are specified at checkout; no trial duration assumed.Free local protection forever. Solo ($4.99/mo) adds cross-device sync and GitHub secret monitoring. Pro ($15/mo) adds real-time alerts and longer retention. Team ($30/seat/mo) adds shared policy and team dashboards. Enterprise plans available with custom terms. Source

Curated by Newtools

Added and checked by the editorial team

Sources

Last verified 8 Nov 2026

More Security

Compare all

Microsoft Defender

Security application that provides always-on protection for personal and family devices against online threats.

Proton VPN

Privacy-focused virtual private network service created by the team behind Proton Mail.

dotenvx

Secure environment variable manager with encryption, auditing, and multi-environment sync.

hCaptcha

Enterprise security platform that provides real-time bot detection, AI agent verification, and fraud prevention.