Try in your browser

The Scaler client runs in your browser on a JupyterLite (Pyodide) kernel that already has the client installed. The scheduler and worker managers still run natively, and the browser can only reach them over a ws:// address.

The quickest way to get a scheduler and worker manager(s) running is the Launchpad: hit Launch, then paste the ws:// address it gives you into a notebook’s SCHEDULER_ADDRESS and run all cells.

Note

Workers must run Python 3.13 to match the in-browser (Pyodide) kernel, with numpy (pinned <2.3 to match Pyodide) and scikit-learn installed on the worker side.

Demo notebooks

Each demo is worker-heavy and client-light: the browser orchestrates a batch of independent tasks while the actual CPU work happens on the workers.

Read the write-up for any example, or launch it straight into the in-browser JupyterLite notebook:

• Parallel square roots (warm-up) — open in browser →
• Heavy object reuse with send_object — open in browser →
• Monte Carlo estimation of pi — open in browser →
• Mandelbrot tile rendering — open in browser →
• Segmented prime sieve — open in browser →
• Word-count map-reduce — open in browser →
• Image batch filter — open in browser →
• Hyperparameter grid search (sklearn) — open in browser →

For heavier real-world notebooks see Examples – these are too heavy for a browser kernel to orchestrate and are best run from a native Python client.