By "maintaining itself" I'm sure the OP means it maintains itself in response to their commands. So nothing changes unexpectedly, or against the user's will. Any changes that prove inconvenient or problematic are fixable.

I should probably RTFA before assuming that, but that's the way my Linux box works. When something breaks or needs upgrading, I just tell the agent to deal with it. Normally that's Claude Code, but the role will be assumed by a local model soon enough.