Compare
Celistra vs SSH + tmux
Quick verdict
SSH + tmux is fine if you have one or two boxes and you live in a terminal. It's not great as a fleet view, has no sandbox story, no mobile approval surface, no audit log, no restart-on-crash. Celistra is what SSH + tmux would look like if redesigned for AI agents.
What SSH + tmux gets right
- Free, ubiquitous, well-understood
- Survives terminal close (tmux owns the session)
- Multiplexing (multiple windows in one session)
What SSH + tmux gets wrong (for AI agents)
- No fleet view. Three boxes = three SSH connections to remember.
- No sandbox per agent. Anything you run has the user's full permissions.
- No mobile-native approvals. Tmux on iSH-like apps is a chore; haptic approvals don't exist.
- No restart-on-crash without scripting. You write your own systemd unit per agent.
- No audit log of "what did the agent do." tmux scrollback is best-effort.
- Auth is per-key. Pairing a new box is the ssh-keygen/ssh-copy-id dance.
Side-by-side
| SSH + tmux | Celistra | |
|---|---|---|
| Persistent session | ✓ | ✓ |
| Fleet view across machines | ✗ | ✓ |
| Phone access | iSH/Termius (clunky) | Native iOS/Android app |
| Haptic approvals | ✗ | ✓ |
| Sandbox per process | ✗ | ✓ |
| Restart-on-crash | Manual systemd | Built-in (5/60s ceiling) |
| 30-day history | tmux scrollback | SQLite, queryable |
| Audit log (verifiable) | ✗ | ✓ Hash-chained |
| Pairing time | 2–10 min (keys, ports, config) | 30s (tray button) |
Use both
SSH still has a place — when you need a real shell on the box for ops tasks, that's SSH. Celistra is for processes, not for terminal access. They coexist.
FAQ
Can I keep using SSH alongside Celistra?
Yes. They don't conflict. SSH for shell, Celistra for process supervision.
Does Celistra encrypt traffic like SSH?
Yes. LAN traffic is loopback (no encryption needed). Off-LAN traffic flows through a TLS-terminated tunnel — Let's Encrypt certs, CAA pinned.