Exposing ports

Run a web app, API or dashboard on your GPU and give it a link, without SSH tunnels.

Expose a port

1. Start your app inside the GPU, listening on 0.0.0.0 (not 127.0.0.1) on a port of your choice, e.g. 8000.
2. Open the GPU, go to the Ports tab and choose Expose port.
3. Enter a name and the port number and pick a type:
   • HTTP gets a public HTTPS link on flint.monster.
   • TCP or UDP gets an address and port to connect to directly.

The link appears straight away. Remove it from the same tab when you're done.

Example: a quick web server

cd /workspace
python3 -m http.server 8000 --bind 0.0.0.0

Expose port 8000 as HTTP and open the link.

Apps behind a path

HTTP links are served under a path like /p/<gpu-id>/8000/. Most frameworks work as they are. If yours builds absolute links (some dashboards do), set its base path to that prefix. The request carries an X-Forwarded-Prefix header with the prefix for apps that read it.

Good to know

• Anyone with an HTTP link can open it. Add your own login if the app isn't meant to be public.
• Exposed ports only work while the GPU is running.
• JupyterLab already has its own Jupyter button; you don't need to expose it.