gradio

Machine learning web interface builder

TLDR

Run a Gradio app with hot-reload (restarts on file changes)

$ gradio [app.py]

Run directly with Python (no auto-reload)

$ python [app.py]

Deploy the current directory as a Hugging Face Space

$ gradio deploy

Deploy with a custom title and entry file

$ gradio deploy --title [my-app] --app-file [app.py]

Print the Gradio environment info for bug reports

$ gradio environment

Run with debug output enabled

$ GRADIO_DEBUG=1 python [app.py]

SYNOPSIS

gradio file [--demo-name name] [--watch-dirs dirs]gradio deploy [--title title] [--app-file file]gradio environment

Gradio is a Python library for building machine-learning demos and web UIs, plus a small companion CLI. The gradio command wraps python to run a script with live reload - whenever the watched files change, the server restarts and the browser refreshes automatically, which is the main reason to use gradio app.py instead of python app.py during development.The CLI also exposes gradio deploy, which packages the working directory and pushes it to Hugging Face Spaces for hosting, and gradio environment, which prints diagnostic information. Server settings such as port, host, share links, and authentication are normally configured via arguments to demo.launch() inside the Python script, or via GRADIO_* environment variables.

PARAMETERS

file

Python file containing a Gradio app. Running gradio file.py launches it with auto-reload, equivalent to python file.py but restarting on edits.

deploy

Subcommand: upload the current directory to Hugging Face Spaces, respecting .gitignore.

environment

Subcommand: print Gradio, Python, and OS version info useful for issue reports.

--title title

(deploy) Name the Hugging Face Space.

--app-file file

(deploy) Path to the entry Python file (default: app.py).

--demo-name name

Name of the demo variable inside the script (default: demo).

--watch-dirs dirs

Additional directories to watch for hot-reload.

ENVIRONMENT

GRADIO_SERVER_PORT

Port to bind (default: 7860).

GRADIO_SERVER_NAME

Interface to bind. Use 0.0.0.0 to listen on all interfaces.

GRADIO_DEBUG

Set to 1 to keep the main thread alive and print stack traces (useful in Colab).

GRADIO_TEMP_DIR

Directory for temporary files such as uploaded media.

PYTHON EXAMPLE

$ import gradio as gr

def greet(name):
return f"Hello, {name}!"

demo = gr.Interface(fn=greet, inputs="text", outputs="text")
demo.launch()

CAVEATS

The gradio CLI is thin: most configuration happens in demo.launch(...) inside your Python script, not via flags. Share links created with share=True expire after 72 hours and tunnel through Gradio infrastructure. gradio deploy requires a Hugging Face token (via huggingface-cli login). Requires Python 3.10+ in recent versions.

HISTORY

Gradio was created by Abubakar Abid and team, acquired by Hugging Face in 2022. It revolutionized ML demo creation by making interactive interfaces accessible to researchers without web development expertise.