We built an SSH client we'd actually trust on a live production call 🎙️

Hey! 👋

Just shipped Thread Nimbus — real terminals (xterm.js + ssh2), but with an AI agent, Devi, that only investigates on its own and needs your explicit approval before touching anything.

The part I'm proudest of: every command — yours or Devi's — gets classified and logged. Destructive stuff like rm -rf / is blocked before it ever reaches a confirm dialog. Plus a one-key "presentation mode" that blanks hostnames/IPs if you're screen-sharing.

Would love feedback from anyone who's had an "oh no" moment pasting secrets into a chatbot.