How to Install and Use Local AI Models with Ollama

In this guide you’ll learn to install and use local AI models with Ollama: a free tool that lets you run artificial intelligence directly on your computer, offline, no subscription, with complete privacy. Your conversations never leave your machine.

Pick your system (Windows PC or Mac) with the buttons above and tap each step to open it.

Before you start: Ollama is 100% free and open source. You need at least 4 GB of RAM for small models and 8 GB or more for medium ones. Speed depends on your hardware, but it works on regular computers.

1 Install Ollama

Go to https://ollama.com/download

Click Download for Windows and download the .exe file.
Open the installer and follow the steps (next, next, install). Ollama installs and runs in the background as a Windows service.
To verify, open PowerShell and type:

ollama --version

If a version number appears like ollama version 0.3.x, the installation was successful.

Click Download for macOS and download the .zip file.
Unzip the file and drag the Ollama app to your Applications folder.
Open the Ollama app (find it in Launchpad or Applications). You’ll see the Ollama icon in the top menu bar.
To verify, open the Terminal and type:

ollama --version

If a version number appears like ollama version 0.3.x, the installation was successful.

2 Download your first model

Ollama downloads the model the first time you request it. Open PowerShellthe Terminal and type:

ollama pull llama3.2

This downloads Llama 3.2 (3B) by Meta, a good starter model: capable, supports Spanish, and only about 2 GB. The download may take several minutes depending on your internet speed.

Other beginner-friendly models (confirm availability at https://ollama.com/library):

Model	Size	Best for
`llama3.2`	~2 GB	General use, good Spanish
`gemma3:2b`	~1.6 GB	Computers with limited RAM
`phi3`	~2.3 GB	Code and reasoning
`mistral`	~4 GB	Long text, good performance

3 Chat with the model

To start a conversation, open PowerShellthe Terminal and type:

ollama run llama3.2

You’ll see the >>> prompt and can type directly to it:

>>> Explain what artificial intelligence is in simple terms.

The model will respond in seconds (or minutes on slower computers). To exit, type /bye or press Ctrl+D.

Useful commands:

ollama list          # see downloaded models
ollama run gemma3:2b # switch to a different model
ollama rm mistral    # delete a model to free up space

4 Choose a model based on your computer

Not all models perform the same on every computer. This table helps you choose:

If you have…	Recommended models
4 GB RAM (CPU only)	`gemma3:2b`, `phi3:mini`
8 GB RAM	`llama3.2`, `mistral`, `phi3`
16 GB RAM or more	`llama3.1:8b`, `deepseek-r1:8b`, `gemma3:12b`
NVIDIA GPU or Apple Silicon	Any model, including large 30 GB+ ones

General rule: download a model whose size is less than half your available RAM. If you have 8 GB of RAM, stick to models of 3–4 GB at most.

Larger models are more capable, but smaller ones are already useful for writing, summarizing, translating, and answering questions.

Shortcut: graphical interface with Open WebUI

If you prefer a visual interface like ChatGPT instead of the terminal, you can install Open WebUI, which connects to Ollama and gives you a full chat with history in your browser. Requires Docker to be installed.

Once you have Docker:

docker run -d -p 3000:80 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data --name open-webui ghcr.io/open-webui/open-webui:main

Open http://localhost:3000 in your browser and you’ll see the interface. Create a local account (only on your computer, nothing goes to the internet) and start chatting.

Open WebUI is also free and open source. Confirm the latest installation instructions at https://github.com/open-webui/open-webui.

We also recommend checking out the tutorial How to Connect an AI API if you want to integrate Ollama into your own applications.

Test that everything works

Open PowerShellthe Terminal.
Type ollama run llama3.2 and wait for the >>> prompt.
Ask it something simple: “What is the capital of France?”
It should answer “Paris” within a few seconds.

If it answers correctly, everything is working! To compare models and find which one fits your needs best, check out the tutorial ChatGPT vs Claude vs Gemini: many of those models are also available locally through Ollama.

If something goes wrong

Problem	Fix
The `ollama` command is not recognized	Restart your computer after installing and try again
The model download gets interrupted	Run `ollama pull model-name` again; it resumes from where it left off
Responses take too long	Try a smaller model (like `gemma3:2b`) or close other apps to free up RAM
The model responds in English even though you ask in Spanish	Start the message with “Always respond in Spanish.” or adjust the language in Open WebUI
”Context length exceeded” error	The model has a text limit per session. Start a new conversation with `/bye` then `ollama run` again

Frequently asked questions

What is Ollama?

Ollama is a free, open-source tool that lets you run AI models directly on your computer, with no internet or subscriptions needed. It works on Mac, Windows, and Linux.

Do I need a powerful GPU to use Ollama?

It's not required, but it helps. Small models (like Phi-3 Mini or Gemma 3 2B) run fine even on CPU-only computers. With a GPU (NVIDIA with CUDA or Apple Silicon) larger models respond much faster.

Are my conversations private with Ollama?

Completely. The model runs on your own machine and your data never leaves it. There are no external servers or company storing your questions.

What models can I use with Ollama?

Hundreds: Llama 3, Gemma 3, Phi-3, Mistral, Qwen, DeepSeek, and many more. All of them are in the official library at https://ollama.com/library. Confirm the updated list on that page.

Can I use Ollama as an alternative to ChatGPT?

Yes, for tasks like writing, summarizing, coding, and text analysis, local models are very capable. For real-time internet search you'll need another tool, since Ollama doesn't have internet access.