Agentinel
Stop AI agents from installing malicious packages.
Start new thread

Agentinel - The local security guardrail for AI coding agents

by•
Because your AI loves slopsquatting. The zero-cost, locally-run firewall for AI agents. Designed to instantly intercept hallucinated and malicious packages across JavaScript, Python, and Rust before your terminal runs them.

Add a comment

Replies

Best
Maker
📌
Hey everyone 👋 When we launched Agentinel, our goal was simple: stop AI coding agents (like Cursor and Claude) from hallucinating and installing malicious `npm` packages on your machine. We built a lightning-fast local firewall, shipped it, and thought we had solved the problem. But we completely overlooked reality. Most of us aren't just writing JavaScript anymore. We're juggling Python scripts and Rust backends in the same project. If an AI agent decided to run `pip install [fake_package]`, our "impenetrable" firewall just let it happen. We were only protecting 33% of the modern stack. In v1.2.0, we just shipped a massive architectural overhaul. Agentinel now intercepts `pip` and `cargo` commands in real time. We had to build custom registry fetchers for `pypi.org` and `crates.io`, and expand our local offline database to handle hundreds of megabytes of Python and Rust vulnerability data—all while keeping interception under 1 millisecond. If you're using AI agents across different languages, you're finally protected across your entire stack. I'm curious—for those of you using AI coding assistants, which ecosystem (JS, Python, Rust, Go) do you see the AI hallucinate packages for the most?