A new open‑source project called Matchlock lets developers run AI agents inside a secure Linux sandbox, protecting the host system from unexpected behavior. It’s designed to be simple to set up and helps keep experimental AI tools safe and isolated. https://github.com/jingkaihe/matchlock