Allowing the agent to operate within a desired operating system is like giving it arms, eyes, and hands in its preferred environment—it’s vital:

the agent has access to files, can read and modify them, create and modify the machine, browse the web or run your apps without restrictions.

In short, we can create an infinite number of use cases by removing permission barriers and granting the AI access to the operating system’s capabilities. The tests we ran were quite impressive: the program runs 10 times faster than in traditional environments like Claudecode, etc. It detects and resolves issues we hadn’t even anticipated in record time. It has access to the environment’s background functions like write_file, read_file, etc. This allows it to write files and use apps in the background without even needing to move.

Furthermore, it removes the security barrier because the agent runs in a dedicated virtual machine; so if a problem arises, we have access to snapshots of our machine’s current state.

Doing this on an unprotected system would be suicidal from a security standpoint.