Ollama lets you download and run AI models on your own Windows, Mac or Linux computer. Install the app, choose a model and start a chat from the terminal or Ollama’s desktop app. Once downloaded, a local model can answer prompts without sending them to a model provider. The practical limit is your hardware: Ollama lists Gemma 4 E2B at about 7.2GB and recommends 8GB of available graphics memory or Mac unified memory for a responsive experience.
A 16GB MacBook Air or a Windows laptop with an 8GB NVIDIA graphics card is a more comfortable starting point. An 8GB office laptop can run smaller models, but replies may take longer, especially if the computer has to use system memory.
What you gain by running AI locally
A local model can be useful when you want to work offline or keep a prompt on your own computer instead of sending it to a cloud AI service. That can help with private notes, code or documents, but check the privacy settings of any app you connect to Ollama. The model running locally does not automatically make every connected app private.
You also avoid paying a cloud service for each request, though you still need to download the model and run it on your own hardware. The trade-off is capability: smaller local models are more likely to struggle with difficult maths, long documents and multi-step coding tasks. They work best as a handy assistant for everyday questions, not as a replacement for every cloud model.
Step 1: Install Ollama
Windows: download OllamaSetup.exe from ollama.com/download. It installs in your user folder without admin rights and needs Windows 10 22H2 or newer. NVIDIA owners should be on driver 551.61 or later. You can also install it from PowerShell:
irm https://ollama.com/install.ps1 | iexmacOS and Linux: download the Mac app, or run the install script in Terminal:
curl -fsSL https://ollama.com/install.sh | shOn Windows and Mac, Ollama continues running in the background after setup. On Linux, start it with ollama serve if the installer did not set up the service, then check the installation with ollama -v.
Step 2: Download and run a model
ollama run gemma4:e2bThe first run downloads the model, so allow for a roughly 7GB download. On a 100Mbps connection that can take about ten minutes, and using a mobile hotspot may consume a sizeable part of your data allowance. Ollama then opens a chat prompt in the terminal. Type /bye to exit. Use ollama pull to download a model without opening it, and ollama list to see what is already on your computer.
If you prefer a chat window, the Ollama app for Windows and Mac includes one. On Mac, you can also connect Ollama models to supported apps such as the ChatGPT and Claude desktop clients from Settings > Apps.
How much RAM do you need?
| Your laptop | What to expect |
|---|---|
| 8GB RAM, no graphics card | Smaller models can run, but replies may be slow. Close memory-heavy apps first. |
| 16GB Mac (M-series) | A more comfortable starting point for models such as Gemma 4 E2B. |
| Windows with 8GB NVIDIA GPU | Good acceleration for compatible models; larger ones may also use system RAM. |
| 32GB+ Mac or 16GB+ GPU | More room to try larger models. |
Leave room on your drive too. Models can take a few gigabytes or much more, and Ollama itself needs about 4GB on Windows. If your C: drive is nearly full, Ollama lets you move the model folder to another drive.
When your laptop is not enough
Ollama also offers cloud models, whose names end in :cloud, such as gemma4:cloud. These run on Ollama’s servers, so they can be used on less powerful hardware, but you need an Ollama account and your prompts leave your computer. If you want more local computing power instead, our NVIDIA PAIR local AI cluster guide explains how to combine machines at home.
Using Ollama with coding tools
Ollama provides an API at http://localhost:11434. Many coding tools can connect to it through OpenAI- or Anthropic-compatible APIs, rather than using a paid API key. The ollama launch command can configure supported integrations. Smaller models may struggle with long coding tasks, but can still help explain a function or draft a short script. If you are connecting coding tools, our guide to MCP servers for Command Code covers another way to extend that setup.
If you already use ChatGPT or Gemini, Ollama does not have to replace them. It gives you another option for offline prompts and small tasks, as long as your computer has enough memory for the model you choose.
- SourceOllama Quickstart
- SourceOllama for Windows
- SourceOllama for Linux
- SourceOllama CLI reference
2 responses to “How to Run AI Models Locally on Your Laptop With Ollama”
[…] Apple is also pushing both machines as good for running AI models locally. Memory matters more than the chip for this, so consider the 24GB or 32GB options if local models are the goal. You can try it on any recent Mac by following our walkthrough on how to run AI models locally with Ollama. […]
[…] A VM like this is a good place to try heavier projects without touching your main computer, such as running AI models locally with Ollama. […]