Mudroom: Run coding agents in a VM, review changes like a PR

Mudroom runs Claude Code, Codex, Gemini CLI and other agents in a Linux sandbox on a copy of your project, behind a network allowlist. Review the diff like a PR and apply only what you want, with undo. Mac app, plus a CLI for Linux and Windows. Free, MIT.

Hi Product Hunt, I'm Karim. I run Claude Code with permission prompts off because it's much faster, and then I'd find out afterwards what it did to my project. Mudroom keeps the agent off your real folder: 1. It makes a copy of the project, untracked and ignored files included. On a Mac that's an APFS copy-on-write clone, so it costs no disk until files change. 2. It starts a Linux sandbox and mounts only the copy. On a Mac that's a micro-VM with Apple's container runtime; on Linux and Windows, Docker or Podman. 3. By default the sandbox has no DNS and no route out. The only exit is a proxy on your machine that allows the agent's API hosts plus whatever you add, and logs every connection. (On Windows this locked mode is still being tested.) 4. When the agent finishes, you review the diff like a pull request and apply all of it, some files or single hunks. Any apply can be undone. It's a native Swift app for macOS 26 on Apple silicon, plus a CLI for Linux, Windows and Intel Macs. MIT licensed. I'd like to hear which agents you'd want tested next.

so .env files end up in the copy too, can you exclude those

Reviewing the diff like a PR, with undo, on a copy of the project behind a network allowlist is the part that stands out to me. Do files the agent creates from scratch, or renames, show up in that same review, or does it only cover changes to files that already exist?

so .env files end up in the copy too, can you exclude those