%401x.png)
We build open-source security for AI agents.
Agents now read files, run commands and call APIs on our behalf. We make sure every one of those actions is checked.
Garmwall is a firewall for AI agents, written in Rust. It sits between an agent and the tools it connects to over the Model Context Protocol (MCP), pins every tool definition, inspects calls and results, and asks a human before anything risky happens.
| Repository | What's inside | Status |
|---|---|---|
garmwall |
The proxy, detectors, policy engine and web dashboard | In development |
.github |
This profile and the community health files shared by every repo | Live |
policies |
Ready-made Cedar policies for popular MCP servers | Planned |
examples |
Setup guides for Claude Code, Cursor, VS Code and Kubernetes | Planned |
- Open source first. The core is and stays Apache-2.0.
- Local by default. Your prompts, code and secrets never leave your machine.
- Honest about limits. We reduce the blast radius; we never promise perfect safety.
- Built in public. Design decisions, roadmap and progress happen in the open.
- Follow the organization and star
garmwallto see progress - Share ideas, questions and use cases in Discussions
- Report security issues privately through the repository's security advisories, never in a public issue
Maintained by Carmine Chessa · Molise, Italy
