Subshell Docs

Tear the instance down from its own machine: what a reset deletes, what it never touches, and the confirmation that gates it.

Reset is the one act that deletes everything the instance has ever written, so it is deliberately kept off the network: no page you serve can drive it, and, decided once and worth stating plainly, subshell-server has no reset command. The sequenced, guarded reset lives in the Subshell Server desktop assistant. A headless instance deletes its data by hand, and this page shows both paths.

From the desktop app

The dashboard's danger card on Settings is only a door: it raises the Subshell Server app's assistant to the reset screen. The remote page can name a screen, never a path, never a command. Everything that follows happens on the bundled assistant page, on this machine.

What the screen shows before you can arm it: every path the reset will delete. They are not a template. The assistant reads them from the server's own status report right now, the same list Files & Paths describes: the database file, the pane logs, the node artifacts, the data directory, and the config home with its config.env. The report is taken all-or-nothing: a partial one refuses to arm at all, because deleting a subset and reporting success is worse than deleting nothing. The plan is held by the app for the whole chain: the sequence stops and uninstalls the very server whose report names the paths, so it cannot be re-asked afterward.

The confirmation is typing this machine's hostname, the same one the screen displayed, compared on the machine itself. Then two guards run before anything is deleted: every target must be an absolute path that is neither / nor your home directory, and any recursive delete that is or contains the installed server binary is refused outright. The binary at ~/.local/bin/subshell-server survives; the config file does not: it is one of the things deleted.

What the chain does, step by step with the screen showing its place: stop the server, close the pane servers (each local subshell's tmux), uninstall the service definition, delete the paths, clear the Subshell Server app's own settings, then restart the app, which re-probes the machine and finds it at first run, ready to set up again.

What no reset reaches:

Enrolled remote nodes. Each keeps its node key and keeps running. It simply stops receiving commands, because the control plane it dialed is gone.
A same-machine node. The subshell daemon and its config belong to the node role, a different program with its own config home, and are untouched. Stop it yourself from Subshell Client, or with subshell service stop on that machine. Subshells running on other machines are not shut down by resetting this one; they live on their nodes' disks and die with their own panes.
The installed binary, by the containment guard above, and the Subshell Client app's own choices.

Backups go too

Database backups are kept inside the data directory, which makes them covered by design: a reset leaves no copy behind. If anything in the database is worth keeping, run subshell-server backup and move the file somewhere else first.

From a headless box, by hand

A headless instance has no assistant window, and there is no command that stands in for one. The by-hand equivalent is deliberately left as hand work so that the only thing deleting your instance is a person at a keyboard on that machine:

subshell-server service uninstall     # stop the service and remove the definition
subshell-server status --json         # the paths block lists everything below

Delete what that report names: the database, the pane logs directory, the data directory, then the config home last (it holds the secret; deleting it is what makes the instance unbootable rather than merely empty). The binary at ~/.local/bin/subshell-server is the only thing the assistant would have kept by policy; removing it is your call:

rm ~/.local/bin/subshell-server       # optional — nothing else removes the binary

Unenrolled leftovers are on the other machines' side: nodes still hold their keys, and there is no plane left to call: deleting them is the node CLI's business on each machine, or nothing's.

Reset

From the desktop app

From a headless box, by hand

See also

On this page