Gradio-Lite: Serverless Gradio Running Entirely in Your Browser
Hugging Face
Hugging Face's Gradio now runs entirely inside your browser, no server needed. It's called Gradio-lite, and it means free, private ML demos that work offline.
Based on reporting by Hugging Face — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Gradio has always been the easiest way to slap a UI on a machine learning model, but it had one nagging requirement: you needed a server somewhere to actually run the Python. Hugging Face just knocked that requirement out with @gradio/lite, a new JavaScript library that runs full Gradio apps inside the browser tab itself, using Pyodide to execute Python as WebAssembly.
The setup is almost comically simple. Drop a script tag and a stylesheet into an HTML file, wrap your normal Gradio Python code in a <gradio-lite> tag, and open the page. No Flask, no Docker, no cloud bill. Want multiple files? Use <gradio-file> tags with a name attribute and mark one as the entrypoint. Need extra packages? A <gradio-requirements> block lets you list them like a requirements.txt, and Gradio-lite installs them via micropip. Hugging Face even shows off a sentiment-analysis app using transformers_js_py that classifies text with a model running fully offline, hosted as a free static Hugging Face Space.
The pitch here is threefold: zero server costs since everything executes on the visitor's machine, snappier interactions because there's no round trip to a backend, and better privacy since user data never leaves the device. For anyone who's ever paid for a GPU instance just to keep a demo alive, that first point alone is a real shift in how cheaply you can distribute an ML app.
It's not free of trade-offs, though. Pyodide itself needs to load before anything runs, which tacks on 5 to 15 seconds of initial wait — annoying for a quick demo, probably fine for something people will actually use. And not every Python package plays nicely with Pyodide's WebAssembly environment, so anything with heavy or unusual dependencies needs a quick compatibility check before you build around it. Hugging Face also put up an interactive playground on gradio.app so people can poke at code and see results without setting up a local file at all.
My take — AI-written commentary, not fact-checked reporting
This is the kind of unglamorous infrastructure work that actually moves the needle for open ML tooling — pushing inference to the client is exactly how you get demos that don't die the moment a hobbyist's free-tier server sleeps. I'd bet within a year half the 'try it live' ML demos floating around Twitter are quietly running on something like this instead of a rented GPU box.
Read more about this at: Hugging Face