sparkDash

What sparkDash does

sparkDash streams GPU, CPU, unified memory, storage, network and local LLM metrics into one browser window. Units can be added, edited, reordered or removed from the UI without restarts. Local hosts are read through sysfs, proc and nvidia-smi, and remote hosts over SSH with a key or password. Any Linux machine with an NVIDIA GPU can be added as a dedicated GPU host, with RAM and VRAM shown separately.

It auto-detects LLM servers including llama.cpp, vLLM and sglang, reports decode and prefill tokens per second, and includes decode, prefill and quality benchmarks. The Hermes Agent integration is opt-in per unit: a background update check every 10 minutes, status badges, and a one-click or batch hermes update. Other opt-in probes cover ComfyUI and Tailnet status.

Key features

• Multi-unit dashboard with live WebSocket metrics for GPU, CPU, memory, storage and network
• Auto-detection of local LLM servers such as llama.cpp, vLLM and sglang
• Decode, prefill and quality benchmarks
• Opt-in Hermes Agent update check, status badges and one-click or batch hermes update
• Non-Spark Linux GPU hosts monitored over SSH with nvidia-smi
• Command palette and four themes

When to use it

• Find which of several Spark units runs an outdated Hermes Agent and update them in one batch
• Benchmark a local model's decode and prefill speed before pointing an agent at it
• Monitor GPU and memory load across a small home-lab of NVIDIA machines

Who it is for: Owners of NVIDIA DGX Spark or other NVIDIA GPU Linux machines who run local models and Hermes Agent and want one monitoring page.

How it fits with Hermes Agent

Hermes Agent support is one opt-in feature of a general GPU and LLM monitoring dashboard, covering update checks and status badges rather than agent control.

Requirements: One or more NVIDIA DGX Spark (GB10) units, or Linux machines with an NVIDIA GPU, reachable locally or over SSH

FAQ