dsh-sysmon

You are deep in a DSH session, prompts flowing, models spinning. You want a casual glance at whether host CPU, memory, or network is spiking, but you do not want to alt-tab to a terminal and run htop, and you certainly do not want to wire up a React monitoring widget on top of the Web GUI. dsh-sysmon fills exactly that gap: a small floating panel in the bottom-right corner that appears the moment you refresh the page, built with zero front-end frameworks on the client side and zero third-party dependencies on the host (just /proc reads and node:os). It polls the same-origin snapshot endpoint every two seconds, pauses automatically when the tab is hidden, and caches snapshots on the host so repeated polls do not re-read /proc. The total footprint is negligible.

The panel covers the metrics that matter day to day: CPU, memory, and disk (multiple physical mount points, root first) each with a progress bar; network download and upload shown on separate rows with auto-scaled bars based on recent peaks. Hovering either network row reveals a per-process TOP 10 overlay, sorted by byte rate when ss provides counters, or gracefully falling back to active connection count when it does not. For GPU, AMD is read directly from sysfs with zero child processes, while NVIDIA samples via nvidia-smi in a non-blocking background loop. The widget is draggable, collapsible to a single line, and hideable with a summon button; position persists in localStorage across refreshes.

It is aimed at developers and operators who are already living inside a DeepSeek Harness process. Linux gives you the full feature set (total network rate, process TOP 10, process count all lean on /proc or ss); non-Linux platforms still get CPU, memory, disk, load, uptime, and GPU, with network and process rows silently degrading to empty rather than breaking anything.