Launched this week
Greywall
Local agent sandbox with real-time network control dashboard
17 followers
Local agent sandbox with real-time network control dashboard
17 followers
AI agents run on your machine, with your credentials, making network requests you never see. Greywall puts a deny-by-default layer in front of any CLI agent. Run greywall -- claude and it restricts file access, blocks outbound connections, and shows you a live dashboard of what the agent is trying to reach. You decide what to allow. Nothing is containerized. Your workflow stays the same but safer. It runs on Linux and macOS and costs nothing.
Nincha
The way we see it: the company selling you the AI shouldn't also be the one defining your security boundary. Greywall is fully independent of any agent vendor, so your trust model stays yours. That's also why we open-sourced it under Apache 2.0.
We've been shipping pretty actively too. The biggest recent addition is credential protection: Greywall now detects API keys and secrets in your environment, replaces them with opaque placeholders inside the sandbox, and injects the real values only at the proxy layer.
Or also reconstructing conversations from HTTP requests, giving the visibility on what's happening under the hood of the agent loop!
Lots more coming. Happy to answer anything here.
Ive been using Greywall for the past month, and I like to think that Greywall is the little snitch of Claude.
The real-time dashboard is a great touch, especially when I'm running Claude code + greywall on exe.dev
I’ve always been concerned about the security of AI agents and never felt fully comfortable running them in my environment. Greywall made me significantly more confident using Claude Code and other CLI agents.
I've been using it with Claude Code, pretty seamless experience.
Hmm one time my Claude Code agent made some API calls that I didn't intend. It looked like the agent was not permitted to make that call but it made it anyway. So I am curious to try this to see if I can avoid surprises again!