How Ollama works / Local Model Notes

What does Ollama actually do?

It's a program that downloads language models and runs them on your computer. Ollama's quickstart says to download it for macOS, Windows or Linux, then open the app or start from a terminal.¹

Everything below comes from Ollama's own documentation, checked on 28 October 2026.² Features change, so confirm on the linked pages.

How do I run my first model?

The quickstart gives one command as its local example, ollama run gemma4:e2b. That one line fetches the model and opens a chat window in the terminal. To leave, you type /bye.¹

The first run downloads the model. Later runs reuse the saved copy.

Where does the model come from? The FAQ says Ollama pulls models from the Internet.² So the first run needs a connection and some patience. The model name after run tells Ollama which one to fetch. Choose one that fits your memory, using the memory estimate guide.

Is there a server running in the background?

Yes. Besides the chat, Ollama runs a small server so other programs can talk to it. The FAQ says it binds 127.0.0.1 port 11434 by default.² That address points back at your own machine.

The same page says you can change the address with the OLLAMA_HOST setting.² The FAQ describes that as exposing Ollama on your network, so ask whoever looks after yours before you try it.

Does the model stay loaded?

For a while. By default, models are kept in memory for 5 minutes before being unloaded.² That means a second question straight after the first is quicker.

If you'd rather free the memory sooner, the FAQ describes an ollama stop command.²

Where do the files go, and how do I check the hardware?

The FAQ lists a models folder for each system, and says you can pick a different folder with the OLLAMA_MODELS setting.² Models are large, so this matters on a small drive.

To see whether a model sits on the graphics card, the FAQ points to the ollama ps command.² The hardware page explains how to read it. On Apple devices, Ollama says it uses the Metal API for GPU acceleration.³

What about the cloud features?

Ollama offers more than local models. Its quickstart says you can sign in to use cloud models, or choose a local model.¹ A cloud model isn't running on your computer, so the privacy picture changes. The privacy page goes through it.