Hey everyone — Michael here from Hashgraph Online.
I wanted to share HOL Guard, an open-source project we’ve been building for locking down what AI agents are allowed to do in practice.
What it is
HOL Guard sits in front of agent tool use (CLI, desktop, MCP servers, etc.) so you can allow, ask, or block risky actions before they run, and keep a record of what happened. It’s local-first, with an optional Guard Cloud option when you want hosted management.
Why it exists
Coding agents and MCP servers often get broad access to shells, files, and secrets. We kept hitting the same gap: useful agents, weak guardrails. Guard is our attempt at a practical control plane for that — not another “AI firewall” pitch deck, just something you can actually install and run.
Links
We’re part of the Vercel Open Source Program. If you’re running agents against real repos or production systems, I’d love feedback — especially on policy defaults, MCP coverage gaps, and what would make this useful in your stack.
Thanks for reading.