A second frontend

The backend owns the project state — sessions, kernels, file watching — and the frontend renders what it is sent. So a desktop and a laptop can attach to the same backend at once and see the same sessions, the same work-state colours and the same REPLs.

Add one

On the second machine, install the frontend role against the same backend SSH alias:

curl -fsSL https://raw.githubusercontent.com/kalidke/ship-of-tools/main/scripts/install.sh | bash -s -- --backend myserver

(On Windows, use the release zip and scripts\install-shortcut.ps1; see Install.) Launch it with sot-launch; it spawns its own ssh child and connects as one more client of the daemon. Nothing on the server changes.

What is shared and what is not

Shared (lives on the backend)	Per frontend
sessions and their agents	window layout and which pane has focus
each workspace's Julia kernel and REPL	the active session and cursor positions
work-state colours and comm messages	the Terminal drawer (a shell on that machine)
files and previews	settings.toml (layout preset, GPU preference)

Closing one frontend affects nothing the other sees: the agents run under the backend's own supervisors, not under any frontend.

Where a session's output lands

With two windows attached, a file an agent shows (sot-fe preview, show-result) reaches the preview pane and the badge of every window. A browser an agent opens (a docs site, a URL, a wglshow figure) opens on one computer: the window you used last, however long ago; with none used yet, the one that connected last. A window counts as used from its second key press, click or wheel step within five minutes, and a switch between windows is noticed within about six seconds. Name a window with --fe <host> to override this. A window that is reconnecting at that moment is not chosen, and a browser open cannot wait for it.

Many machines, one workspace

To make a server reachable from every frontend in a fleet — not only the one you installed it from — enrol it on the hub: see Enrolling a host.