# HOL Guard

> Local approvals for AI coding agent actions before they touch your machine

- Category: [Coding AI](https://finderatlas.com/categories/coding-ai)
- Pricing: Freemium
- Website: https://hol.org/guard
- Platforms: macOS, Linux, CLI
- Made by: Hashgraph Online DAO LLC
- Last updated: 2026-10-25
- FinderAtlas page: https://finderatlas.com/tools/hol-guard

## Overview

HOL Guard is a local-first runtime security layer for AI coding agents such as Claude Code, Codex, Cursor and OpenCode. It runs on your own machine next to the agent and applies a policy at supported agent events. When the agent tries a risky shell command, reads a secret such as a .env file, calls an MCP tool or installs a package, Guard can allow, observe, ask or block the action and keeps a local record of what happened.

## Key features

- Pre-action interception
- Policy outcomes
- Local audit trail
- MCP, skills and plugin inspection
- Layered policy
- Plugin scanner
- Risk check
- Coverage matrix
- Broad agent support

## Pricing

- **Free** ($0): Local protection on one machine, no account required
- **Solo** ($4.99/month): Personal Cloud memory across two devices
- **Pro** ($15/month): Longer history and alerts for one person
- **Team** ($30/user/mo): Shared policy and fleet visibility for teams
- **Enterprise** (Custom): Identity review, evidence export and self-hosted deployment review

## Best for

- Approving risky agent actions locally, Protecting secrets from coding agents, Reviewing MCP servers, skills and plugins before use

## Who uses it

- Software developers
- Security engineers
- Engineering teams
- Plugin and MCP server maintainers

## Strengths

- Open source under Apache-2.0 with public code and release history
- Works offline with no account on the free local tier
- Lets you approve or block actions instead of only logging them
- Covers MCP servers, skills, plugins and package installs, not just shell commands
- Publishes a coverage matrix and states its limits openly

## Limitations

- Coverage depth varies by agent and event type, and some integrations are alpha
- Install instructions cover macOS and Linux; Windows support is not stated
- Cloud tiers add cost, and annual prices are not listed
- Site has no visible terms or company address on the product page

## Not recommended for

- Network-level filtering
- Users who want a guarantee against prompt injection

## Integrations

- Claude Code
- Codex
- Cursor
- OpenCode
- GitHub Copilot CLI
- GitHub Actions
- MCP servers

## FinderAtlas editorial scores (out of 10)

- Action Interception: 8/10
- Agent Coverage: 7.8/10
- Ease of Setup: 7.5/10
- Transparency & Openness: 8.6/10
- Pricing Clarity: 7.2/10

## Alternatives

- [OpenRouter](https://finderatlas.com/tools/openrouter): One API and one bill for hundreds of AI models across many providers ([compare](https://finderatlas.com/compare/hol-guard-vs-openrouter))
- [OpenAI Codex](https://finderatlas.com/tools/openai-codex): OpenAI's coding agent that writes, reviews and runs code from terminal, IDE, desktop or cloud ([compare](https://finderatlas.com/compare/hol-guard-vs-openai-codex))
- [Groq](https://finderatlas.com/tools/groq): Fast, OpenAI-compatible inference API for open models, running on custom LPU chips ([compare](https://finderatlas.com/compare/groq-vs-hol-guard))
- [Replicate](https://finderatlas.com/tools/replicate): Run, fine-tune and deploy open-source AI models through a pay-per-use API ([compare](https://finderatlas.com/compare/hol-guard-vs-replicate))
- [Langfuse](https://finderatlas.com/tools/langfuse): Open-source platform to trace, evaluate and improve LLM applications and agents ([compare](https://finderatlas.com/compare/hol-guard-vs-langfuse))
- [CodeRabbit](https://finderatlas.com/tools/coderabbit): AI code review that checks, explains and prioritizes every pull request before merge ([compare](https://finderatlas.com/compare/coderabbit-vs-hol-guard))
- [Greptile](https://finderatlas.com/tools/greptile): AI code review that checks every pull request against your full codebase context ([compare](https://finderatlas.com/compare/greptile-vs-hol-guard))
- [Warp](https://finderatlas.com/tools/warp): Open-source terminal with built-in AI agents and cloud infrastructure for coding workflows ([compare](https://finderatlas.com/compare/hol-guard-vs-warp))

## Recent updates

- 2026-10-24 (v3.56.0): Latest release listed on the product page, which highlights per-app protection settings

## FAQ

**Is HOL Guard free?**

The local tool is free and open source under Apache-2.0 and needs no account. Guard Cloud is optional and costs $4.99 (Solo), $15 (Pro) or $30 per seat (Team) per month, with Enterprise on request.

**Which AI agents does it support?**

The site lists Codex, Claude Code, Cursor, OpenCode, Antigravity CLI, Hermes and OpenClaw, plus GitHub Copilot CLI and several others. Coverage depth differs by agent, so check the public coverage matrix.

**Does it work offline?**

Yes. Guard Local applies policy and keeps its decision trail on your machine without internet access or a cloud account.

**Does it stop all prompt injection?**

No. HOL states Guard is not a complete prompt-injection preventer and not a network firewall. It controls supported actions before they run.

**How do I install it?**

Run pipx install hol-guard and then hol-guard init. The package is published on PyPI and requires Python 3.10 or newer.

**Who makes it?**

HOL (Hashgraph Online), which publishes the code under the hashgraph-online GitHub organisation. The legal entity listed on hol.org is Hashgraph Online DAO LLC.