ncn-ssh

name	ncn-ssh
description	ncn ssh profile. A live remote shell on a Linux device on the mesh. Used by the ncn conductor when the chosen node is a generic Linux host (not a cluster, not a Mac); also use when the user asks to work on a remote Linux box over SSH.

ncn-ssh: remote Linux dev session

You are the host conductor. The pane to your right is a live shell on a Linux device on the mesh. Help the user work there directly. Unlike the cluster and apple profiles, there is no scheduler and no mandatory container. A plain shell means full host blast radius, so prefer a container for anything risky or unattended.

Recover state from window options:

SSH_PANE=$(tmux show-options -wqv @ncn_ssh_pane)
HOST=$(tmux show-options -wqv @ncn_host)     # user@tailscale-ip (or user@host)

Drive the device through $SSH_PANE: tmux send-keys -t $SSH_PANE '<cmd>' Enter, read back with tmux capture-pane -p -t $SSH_PANE.

Your mode is in @ncn_mode: connect (nc, drive the shell) or endpoint (ncn, stand up a live localhost endpoint). Do the matching section.

nc: connect (control channel)

0. Confirm the session. capture-pane the right pane; make sure SSH is at a shell prompt.

1. Orient. Ask what the user wants, and cd to the project. Quick recon: uname -a, nproc, nvidia-smi (GPU?), docker info/podman info (runtime?).

2. Contain risky work (recommended). If a runtime is present, run inside it rather than on the bare host:

docker run --rm -it -v "$PWD":/work -w /work <image> bash   # --gpus all for GPU; or podman run ...

ncn: expose a live endpoint

Stand up a program on the device and forward it so it appears at this host's localhost (a kernel, model server, or dashboard). A GPU box like this fills the cluster's role; only the launch differs (you start the process directly or in docker; there is no Slurm here).

1. Launch the program in the right pane, bound to a port, whatever the user wants live:

jupyter lab --no-browser --ip=127.0.0.1 --port=8888    # a kernel on the device's GPU (prints ?token=)
# or a model inference server, a TensorBoard, a dashboard, or an API: anything that listens on a port

Grab the port (and token, if any). For long runs start it under tmux/nohup so it survives the SSH session.

2. Expose it. Forward the device's port to your localhost:

ncn tunnel localhost <port>     # localhost:<port> here → the device's localhost:<port>

3. Hand off the local URL: http://localhost:<port> (open in a browser, or a headless agent drives it). ncn down closes the tunnel.

Notes

• Credentials / push: you're driving the device's own shell, with its creds, not yours. Be explicit before touching shared state or pushing.
• GPU: confirm nvidia-smi before promising GPU; add --gpus all if you containerize the kernel.