A hands-on sandbox. This page is a Static Space — yet it runs a real machine-learning model live, right here in your browser. Scroll down to try it.
No server. The DistilBERT model (~67 MB) is streamed from the HF Hub CDN and runs via Transformers.js (WebGPU, falling back to WASM). First run downloads the model; after that it's cached.
Every Space is just a git repo. A README.md
frontmatter block configures the runtime — the SDK, the hardware, the title. That's it.
--- title: Spaces Capabilities Demo emoji: 🧪 sdk: static # gradio | docker | static pinned: false ---
This very Space uses exactly that. Change sdk: and you change
what kind of app HF builds and serves for you.
Pick the SDK that fits the app.
Python. The most common Space. A few lines of gr.Interface(...) gives you
a UI + an auto-generated REST/MCP API. Supports ZeroGPU.
Any container, any stack. Streamlit, FastAPI, Node, Rust… full control via a
Dockerfile. For when you outgrow Gradio.
Plain HTML/JS, or a built React/Vue/Svelte app. In-browser ML with WebGPU/WASM. Free for everyone — you're looking at one now.
free · no hardwareStatic needs none. Gradio & Docker run on compute and generally need a paid plan — with one big exception (ZeroGPU) for free accounts.
| Tier | What it is | Free account? |
|---|---|---|
| Static | No server; browser does the work | ✓ unlimited |
| ZeroGPU | On-demand A10G-class GPU, per request | ✓ up to 2* |
| cpu-basic | 2 vCPU / 16 GB, always-on | ✗ paid plan |
| Dedicated GPU | T4 → H200, billed hourly | ✗ paid plan |
* Free personal accounts in good standing (verified email, > 30 days old) can host up to 2 ZeroGPU Gradio Spaces. Newer accounts can request a community grant.
A Gradio Space is also a REST API and an MCP server — every function becomes a callable tool.
Inject API keys as hidden secrets, or public config as env vars, at creation or later.
Spaces are stateless, but mount a bucket at
/data for durable storage.
Every Space embeds as an iframe, and Gradio Spaces expose a JS/Python client.
Proxy to Cerebras, Together, Fireworks… to run huge models with zero local VRAM.
Likes, discussions, duplication — anyone can fork your Space with one click.