Ollama Scout: Testing for Exposed Endpoints

Leaving an Ollama API endpoint wide open to the internet is a recipe for disaster, especially if you're running heavy models on your own hardware and don't have an auth layer in place. I've been looking into how easy it is to discover these exposed instances, and Ollama Scout is a pretty direct way to audit this.

The core issue is that by default, many users bind Ollama to 0.0.0.0 to allow remote access but forget to implement a reverse proxy or a VPN for security. This essentially turns your GPU into a free public API for anyone who can find the IP.

If you're trying to verify if your own deployment is leaking, here is the basic workflow to check your exposure:

1. Use a scanning tool like Ollama Scout to probe for the default port (11434).
2. Attempt a basic API call to see if the instance responds without a key.
3. Check the /api/tags endpoint to see which models are currently loaded.

For those who found their endpoints are exposed, you should immediately move to a more secure AI workflow. A quick fix is to bind the service back to 127.0.0.1 or set up an Nginx reverse proxy with basic auth.

Here is the repository for the tool if you want to run a deep dive on your own network:

https://github.com/Ember2819/ollama-scout