Ollama is a free program you install on your own computer to download AI models and run them locally, meaning your questions are not sent to an online service. All you need is one installation, one command and a few minutes of downloading.
This guide, updated for November 2026, walks you through Windows, Mac and Linux: from installation to a small first model, and then how to stop it and free up disk space. The commands come from the official documentation.
What is Ollama, and do you have to pay?
Ollama is a program that downloads and runs AI models on your computer. A "model" is the file that holds the AI's brain: once downloaded, you can chat with it without connecting to outside services. According to the documentation, local models need no API key.
Ollama also offers "cloud" models, which work with an account and are a different thing: you do not need them for this guide. For any paid plans, check the official site. Each model also has its own license, shown on the model page.
How do you install Ollama on Windows, Mac and Linux?
| System | How to install | Stated requirements |
|---|---|---|
| Windows | Download OllamaSetup.exe from ollama.com/download, or use the PowerShell command below | Windows 10 22H2 or later; the installer does not need administrator rights |
| Mac | Download Ollama.dmg from ollama.com/download, or use the Terminal command below | macOS Sonoma (14) or later; Apple M chip or Intel (CPU only on Intel) |
| Linux | Use the terminal command below | No system requirements listed on the download page |
On Windows, open PowerShell and paste this command:
irm https://ollama.com/install.ps1 | iexOn Mac and Linux, open a terminal and use this one:
curl -fsSL https://ollama.com/install.sh | shFor Windows, the documentation states at least 4 GB of space for the program, plus space for models, which can range from tens to hundreds of GB. After installation Ollama keeps running in the background, and you can open the app or type ollama in a terminal.
How do you try your first model?
With Ollama installed, open a terminal (PowerShell on Windows) and start a small model. The first time, the model downloads automatically, then the chat starts:
ollama run gemma3:1bType your question and press Enter. To leave the chat, type /bye. If you prefer a graphical interface, the Ollama app lets you pick a local model and chat without using the terminal.
Which model should you start with?
On an ordinary PC, start with a small model. In the official library, the gemma3 family (from Google) offers these versions, with the download sizes shown on the page in November 2026:
| Model | Download | Context | Input |
|---|---|---|---|
| gemma3:270m | 292 MB | 32K | Text |
| gemma3:1b | 806 MB | 32K | Text |
| gemma3:4b | 3.4 GB | 128K | Text, images |
The practical advice is gemma3:1b for a first test: it is light and downloads quickly. If your PC can handle it and you want better answers, move up to gemma3:4b. The "context" is how much text the model can keep in mind in one conversation. The library also has other small families, such as llama3.2 (1b and 3b) and qwen3 (0.6b, 1.7b, 4b): each page lists names and sizes.
Which commands do you really need?
| Command | What it does |
|---|---|
| ollama run model-name | Starts the model and opens the chat (downloads it if missing) |
| ollama pull model-name | Downloads a model without starting it |
| ollama ls | Lists the downloaded models |
| ollama ps | Lists the models currently running |
| ollama stop model-name | Stops a running model |
| ollama rm model-name | Deletes a downloaded model |
How do you stop the model?
To leave the chat, use /bye. The model stays loaded in memory for 5 minutes after the last use, though (the default stated in the FAQ). To unload it right away:
ollama ps
ollama stop gemma3:1bThe ollama ps command also shows, in the PROCESSOR column, whether the model runs on the graphics card, the processor, or split between the two. To shut Ollama down completely, on Windows quit the app from the notification area icon; on Linux with the systemd service, use:
sudo systemctl stop ollamaWhere are the files saved, and how do you free up space?
- Windows: C:\Users\username\.ollama\models
- Mac: ~/.ollama/models
- Linux: /usr/share/ollama/.ollama/models
These are the default paths stated in the FAQ. You can move them with the OLLAMA_MODELS environment variable. To free up space, do not delete folders by hand: list the models and remove them with ollama rm.
ollama ls
ollama rm gemma3:1bTo uninstall Ollama on Windows, use Settings, Apps; for Mac and Linux the documentation lists the folders to remove. Note: if you changed the models folder, the Windows uninstaller does not delete the models.
What are the most common errors?
- Model too big: if the model does not fit in the graphics card memory, it may run partly on the processor and be very slow. Check with ollama ps and switch to a smaller model.
- Low memory: for gemma4:e2b (about 7.2 GB to download) the documentation recommends 8 GB of VRAM or unified memory on a Mac; with less it works but may be slower. On a PC with little memory, start with gemma3:1b or gemma3:270m.
- Low disk space: models take up gigabytes. Check free space before downloading and remove the ones you do not use.
- The command does not respond on Linux: if the service is not running, the documentation says to start it with ollama serve.
- Wrong model name: copy the exact name, with the version after the colon, from the model page in the library.
What changes for people who work?
A local model lets you try AI on texts you do not want to send online, with no subscription and even offline once downloaded. Small models are less capable than the big online ones: they suit tests, short summaries and drafts, and are weaker for complex tasks. Before using a model for work, read its license on the library page.
Frequently asked questions
Do you need a graphics card to use Ollama?
It is not stated as mandatory. The documentation mentions acceleration on NVIDIA and AMD cards (and on Apple M chips), but Intel Macs use only the CPU. Models run slower on a CPU, so choose small ones.
How much space do you need?
For Windows the documentation states at least 4 GB for the program plus the models: for example gemma3:1b is 806 MB and gemma3:4b is 3.4 GB.
Can you use Ollama without internet?
You need internet to download the model. After that, local requests go to localhost and need no key.
Official sources
- ollama.com/download
- docs.ollama.com/quickstart
- docs.ollama.com/cli
- docs.ollama.com/faq
- docs.ollama.com/windows
- ollama.com/library/gemma3
This article was prepared with the help of artificial intelligence tools, starting from the official documentation, and is reviewed before publication.













