wtf is spark-studio?

Spark Studio is a web based dashboard for running and managing AI models on DGX Spark hardware, a small dedicated machine built for running large language models locally. The idea is to give you one friendly control panel instead of juggling several command line tools. From the dashboard you can launch a local model with a single click, watch memory usage and logs update live, chat with the model once it is running, and benchmark how fast it responds. It supports several different engines for actually running the models, including vLLM, SGLang, llama.cpp, and a browser based option called WebGPU, along with community made setup recipes through a tool called sparkrun. One of the more unusual features is that when something goes wrong, or a model is running slower than it should, you can ask Claude Code or Codex, both AI coding assistants, to look at the logs, figure out the problem, and automatically fix and relaunch the setup for you. This uses your own existing subscription to those tools rather than a separate paid API key. There is a similar automatic mode for speed: it benchmarks a working setup, tries different tuning options, and keeps whichever configuration actually measured fastest. The dashboard also includes a first time setup wizard that checks your system and recommends a model suited to your hardware, a library for saving and sharing configuration recipes, and support for running the interface from any device on your local network, though it warns that there is no login built in, so it should not be exposed directly to the internet. Getting started is done with a single install script, or by cloning the repository and running a start script, after which the dashboard opens in your browser. The full README is longer than what was shown.