dsh-top

When you are deep in a DeepSeek conversation through the DSH web GUI, checking whether the server is running hot means alt-tabbing to a terminal and firing off top or htop — a small friction that pulls you out of the flow. dsh-top pins a live system monitor to the top-right corner of the browser window: real-time CPU utilisation with core count, memory and disk usage shown as percentage bars, network upload and download throughput, and a Top 6 process table with PID, name, CPU %, and MEM %. The panel is colour-coded in a monospace face, draggable by its title bar, and collapsible with a minus/plus toggle; it polls every two seconds while visible and pauses the moment you fold it away.

CPU and memory meters are guaranteed on every platform because they rely solely on Node's built-in os module with no subprocess involved. Disk, network, and process collectors are adapted per OS — df, /proc/net/dev, and ps on Linux; netstat and BSD ps on macOS; PowerShell applets on Windows — and in containerised environments the code walks the cgroup hierarchy to report the container's actual memory limit and CPU quota rather than the host's totals. On slim or distroless images missing those tools, the affected sections simply render as n/a while the rest of the panel keeps working, and every external command runs with a static argument array and a five-second timeout so a stuck collector degrades gracefully instead of hanging the request.

If you use DeepSeek through the DSH web interface on a local machine, a remote server, or inside Docker and Kubernetes, dsh-top lets you glance at the top-right corner between prompts to confirm how much headroom is left and which process is burning cycles — no window switching, no memorised commands.