Frederik von der Heyden

Intelligent Systems

GuardRail

AI coding agents execute commands on your server. GuardRail checks every command before execution and blocks dangerous operations. Open source, born from real production incidents.

Case Study

!

Problem

AI coding agents have shell access to production servers. A single hallucination can wipe a database, leak secrets, or force push to the production branch. Existing solutions log after the damage. None check before.

Solution

Pre-execution hook system that checks every command against a guard chain. Dangerous commands are blocked, safe ones pass without delay. No dashboard, no YAML, no SaaS dependency. One command installs everything.

Result

The system runs in our own production across 15 applications. 41 regression tests continuously validate the guard chain. Available as an npm package and on GitHub.

What the solution does

01

Pre-Execution Guards

Every command is checked before it runs. Not after. No logs, no alerts after the damage. The dangerous operation is prevented.

02

From Real Incidents

Every guard exists because something went wrong in production. Database wipe, secret leak, firewall disabled, force push to production branch. No theoretical threat models.

03

EU AI Act Tooling

Article mapping, risk classification per guard, audit trail for every blocked command. Helps document your governance efforts.

04

Open Core

10 core guards are MIT licensed and free. 48 Pro guards for advanced threat patterns. Both tiers run the same architecture.

Key figures

58

Guards

10

Core (free)

48

Pro Guards

41

Regression Tests

Tech stack

BashClaude Code HooksnpmExpress.jsStripeSQLite

Security for your AI agents?

GuardRail helps protect AI coding agents from dangerous operations. Open source, installed in 30 seconds.

Get in touch

Questions, ideas, or a specific project?

Get in touch