The control plane stays one process (houdry serve). Each extra GPU box is another worker.

On the serve machine, mint a one-time enroll token:

houdry node enroll create

On the new workstation:

curl -fsSL https://github.com/houdry-genomex/houdry/releases/latest/download/install.sh | sh
export PATH="$HOME/.houdry/bin:$PATH"
houdry gpu detect
houdry gpu register

Windows: install.ps1, then houdry.exe gpu detect and gpu register. Use the HDRY_… token when register enrolls. Do not pass it as --token.

If discovery fails:

houdry gpu register --server https://HOST:8080

Leave register running. Ctrl+C drains, waits for the current job, then leaves. Same command as houdry node join. Do not use houdry gpu join.

Also install Ollama on that box and pull at least one model.

Agent does not need a reinstall. model=auto will start using the new READY node when it scores better for a task.

Check:

houdry node list --server https://127.0.0.1:8080