The app always runs on the Spark. Your browser can be on the Spark, on a laptop on the same Wi‑Fi, or on a machine joined to a tailnet. The only thing that changes is the URL and whether you need a tunnel.

Tools in DGX Lab assume local disk and Spark-class memory. Defaults match a stock DGX Spark layout; everything below can be overridden with environment variables read in backend/app/config.py.

DGX Lab expects to run on the DGX Spark (or at least on a box where nvidia-smi and your model cache match how the tools query the system). This is not a hosted product: you clone, you run, you own the outcome.

DGX Lab is a local-first developer dashboard for the NVIDIA DGX Spark. Eight tools for model management, experiment tracking, agent observability, GPU profiling, training recipes, synthetic data, data curation, and dataset browsing -- all memory-aware against 128 GB of unified LPDDR5X.

This post walks through what it is, what it runs on, and how it fits together.