Local models with Ollama: an AI that lives on your computer

Ollama is a program that lets an ordinary computer run a genuinely capable AI model entirely on its own — no account, no monthly fee, no company watching what you type. This is the foundation the rest of this site builds on.

What "local model" actually means

Every AI chatbot needs a model — the trained program that reads your question and writes a response. Most popular chatbots run their model on a distant company server: you type on your screen, your words travel over the internet, a huge computer somewhere answers, and the reply travels back.

A local model skips all of that. The model itself — often a few gigabytes of data, roughly the size of a couple of movies — sits on your own hard drive. Your laptop or desktop does the thinking. Nothing is sent anywhere, and it works the same with the Wi-Fi turned off.

Will it be as good as the famous AI chatbots?

Honestly, for the biggest, hardest problems — no, not quite. But for everyday writing help, summarizing, brainstorming, and answering questions about things you already know, a local model running on a normal modern computer is more than capable.

Getting started, step by step

Check your computer. Ollama runs on Windows, Mac, and Linux. With 8GB of memory you can run a small model comfortably; 16GB or more opens up noticeably better ones. If you have a Mac with an M-series chip, or a desktop with a dedicated graphics card, you're in good shape.
Download Ollama. Visit ollama.com and download the installer for your operating system, the same way you'd install any other program.
Install it normally. Click through the installer as you would for any application — there's nothing unusual about this step.
Pull your first model. Ollama uses simple commands typed into a plain text window. Rather than name a specific version here — they are replaced every few months — go to ollama.com/library and pick a recent release from one of the well-established families: Llama, Qwen, or Gemma for general use, or Phi and the smallest Gemma builds if your computer is older. Choose the smallest size in the family first. You can always download a bigger one later.
Start talking to it. Once downloaded, you can chat with the model directly, or connect it to a friendlier visual interface (see below) if typing commands isn't your idea of a good time.

Prefer a screen with buttons, not commands?

You're not alone, and you don't have to use the command window at all. Free interfaces like Open WebUI or LM Studio give Ollama a proper visual chat window — type in a box, get a reply, just like the AI chatbots you may have already tried, except everything stays on your computer.

What a local model is genuinely good for

Drafting letters, emails, and notes in your own voice.
Explaining a confusing document in plain language.
Brainstorming — trip ideas, gift ideas, project plans.
Practicing a conversation or rehearsing what to say before a difficult call.
Working with the private document search described in Chapter 01.

What to expect the first time

The first download can take a little while depending on your internet connection — the model itself is a real file, not a link to a website. After that, everything works offline. The first few words of a reply may take a moment to appear while the model loads into memory; after that it types along steadily. Responses may feel slightly less polished than a famous cloud chatbot, and that's a fair trade for something that answers to no one but you.