abshkbh/arrakis ? reverse-engineered prompt

Build me a self hosted sandbox service for AI agents that can safely run untrusted code on a Linux machine. I want to be able to start a fresh sandbox, upload files, run commands, and then either destroy it or save a snapshot and restore it later so an agent can backtrack to an earlier state exactly.

It should expose a simple REST API and a small command line tool, and it would be great if a Python SDK can talk to it too. Each sandbox should boot into Ubuntu with a code execution service and a VNC desktop already running, including Chrome for computer use. Please also handle port forwarding automatically so I can connect to the GUI without extra setup.

Keep it secure with MicroVM isolation, and make the setup as straightforward as possible on a self hosted Linux server. If you need to check current docs or library details online, go ahead and do that.