wtf is sparkdash?

sparkDash is a web dashboard for monitoring one or more NVIDIA DGX Spark machines in a single browser window. The DGX Spark is a small but powerful ARM64 computer built around NVIDIA's GB10 chip, designed for running local AI models. sparkDash streams live metrics from these machines, covering GPU usage, CPU load, memory, storage, network activity, and local language model performance. You can add, edit, reorder, or remove Spark units directly from the interface without restarting anything. The main features center on real-time visibility and control. A shared Overview page gives you a bird's-eye look at all connected Sparks, while each unit gets its own tabbed detail page. The dashboard can connect to Sparks running locally or remotely over SSH, using either a key file or a password. For language model workloads, it automatically detects common serving software like llama.cpp, vLLM, or sglang, and reports live tokens per second. It also includes a prompt showcase mode that streams up to 32 concurrent prompts across multiple terminals, and a decode benchmark for measuring throughput under different concurrency levels. Security and configuration are handled with care. SSH passwords are encrypted using AES-256-GCM and are never written to the main config file or returned through the API. The system supports assigning roles to each Spark, such as Head, Worker, or Standalone, with worker units linking back to their head node. You can also trigger graceful shutdowns or Wake-on-LAN power commands, and batch actions are available from the Overview page. The project is built with React 19 and TypeScript on the frontend, running on Express 5 with WebSockets on the backend. It is designed to run inside a single Docker container, which is the recommended way to deploy it. You can also run it in a development mode with hot reloading. Four themes are available, and your choice persists across sessions. The full README is longer than what was shown.