The short version: an SSH client that trusts you with real terminals and your own keychain and an AI agent that has to earn your trust, one command at a time.
No engagement bait. No "AI can do anything now." Just an agent that investigates, explains, and asks before it acts.
Real terminals, secrets that never leave your OS keychain, and an optional AI agent that investigates your servers over your live session but can't run a destructive command, and needs your OK for everything else.