about csvfit

A browser-only machine learning playground. Drop a CSV, train a model, inspect the results. No accounts, no uploads, no backend.

How it works

Everything runs inside a Web Worker in your browser. A Python runtime (Pyodide) compiled to WebAssembly loads AutoGluon and scikit-learn directly into the tab. When you drop a CSV, the bytes stay in browser memory; training happens on your CPU; the leaderboard and evaluation plots render from local state.

There is no server to send data to. The site is a static export hosted on a CDN. Open your browser's DevTools network panel while training — you will see the Pyodide runtime and the snapshot tarball load once, and nothing else.

Why

Most “upload your data” ML tools are thin wrappers around a server that keeps a copy of what you send. For quick exploration on a dataset you would rather not hand to a third party, the friction is real. csvfit is the version where the browser is the compute and the data never moves.

What it is not

Not a replacement for a proper training pipeline. The browser caps memory at a few hundred megabytes and runs on a single thread. Current models: KNN. Good for learning, prototyping, and datasets that fit comfortably in RAM. Not good for gigabyte-scale data or production training.

Contact

Questions, bug reports, or consulting inquiries: