How to Run an AI Model on Your Own Laptop (Free, Private, Offline)
Install Ollama, pick a model that fits your RAM, and chat with an AI that never sends your data anywhere. Takes about 15 minutes.
You can run a capable AI chatbot on an ordinary laptop, for free, with nothing leaving your machine. The easiest way is Ollama. Install it, run one command, and you're chatting with a model like Google's Gemma or Alibaba's Qwen.
The only real decision is which model, and that depends almost entirely on how much memory your computer has. There's a table for that below.
What you need
- A Mac, Windows PC or Linux machine from the last few years
- At least 8 GB of RAM. 16 GB makes a big difference.
- 5 to 20 GB of free disk space, depending on the model
- An internet connection for the first download only
A graphics card helps but isn't required. Apple Silicon Macs (M1 and later) are especially good at this, because the processor and graphics share the same memory.
Step 1: Install Ollama
Go to ollama.com/download, download the version for your system and open it like any other app. On Mac and Windows that's it. Ollama runs quietly in the background.
To check it worked, open a terminal (Terminal on Mac, PowerShell on Windows) and type:
ollama --version
If you see a version number, you're ready.
Step 2: Pick a model that fits your memory
This is where most people go wrong. A model has to fit in your RAM with room left for everything else you have open. If it doesn't, your computer starts swapping to disk and responses slow to a crawl.
A simple rule: take the model's download size, add 2 to 3 GB for your browser and system, and make sure the total is under your RAM.
| Your RAM | Good model to start with | Download size | Command |
|---|---|---|---|
| 8 GB | Qwen 3.5, 4B | 3.3 to 4 GB | ollama run qwen3.5:4b |
| 8 GB | Gemma 3, 4B | about 3.4 GB | ollama run gemma3:4b |
| 16 GB | Qwen 3.5, 9B | 6.6 to 7.6 GB | ollama run qwen3.5:9b |
| 16 GB | Gemma 4, 12B | 7.7 to 8 GB | ollama run gemma4:12b |
| 32 GB+ | gpt-oss, 20B | larger | ollama run gpt-oss:20b |
| 32 GB+ | Qwen 3.5, 27B | 17 to 20 GB | ollama run qwen3.5:27b |
The "B" is billions of parameters. Bigger usually means smarter and slower. On an 8 GB laptop, close your browser tabs before you start; a 4B model with breathing room beats a bigger one that's gasping for memory.
Both Qwen 3.5 and Gemma 4 can also read images. Drag a screenshot into the chat and ask about it.
Step 3: Start chatting
Run the command for the model you picked, for example:
ollama run qwen3.5:4b
The first time, Ollama downloads the model. After that it starts in seconds. You'll get a >>> prompt. Type a question and press Enter.
A few useful commands inside the chat:
/byeto quit/clearto start a fresh conversation"""to paste text that spans several lines (type it again to finish)
Outside the chat:
ollama listshows the models you've downloadedollama rm qwen3.5:4bdeletes one to free spaceollama pull gemma4:12bdownloads a model without opening a chat
Step 4 (optional): Get a proper chat window
The terminal works fine, but if you'd rather have something that looks like ChatGPT, install Open WebUI or LM Studio. Both connect to models on your machine and give you chat history, file uploads and a model picker. LM Studio is the easier install for non-developers. Open WebUI is better if you want to share your local AI with others on your home network.
What local models are good at (and what they're not)
Be realistic about what a 4B or 9B model can do. They're surprisingly good at:
- Summarizing documents and emails you'd never paste into an online service
- Rewriting and proofreading text
- Explaining code and writing small functions
- Brainstorming and drafting
- Working with no internet at all, on a plane or a train
They're noticeably weaker than the big online models at long, multi-step reasoning, at recent events (they only know what was in their training data) and at very long documents on low-memory machines.
If you need the big online models for coding, we compare the paid options in Copilot, Cursor or Claude Code?.
The trade-off is the whole point, though. Nothing you type leaves your computer, there's no monthly bill and nobody can change or remove the model you're using.
Troubleshooting
It's painfully slow. The model is too big for your RAM. Try the next size down, and close other apps.
"Model not found" error. Check the exact name on the model's page at ollama.com/library. Names change as new versions come out.
The fan is very loud. That's normal. Running a model uses your processor hard, so expect heat and some battery drain on laptops.
FAQ
Is Ollama free?
Yes. Ollama and the models in this guide are free to download and run on your own computer.
Is it really private?
Models run entirely on your machine. After the download, you can disconnect from the internet and everything still works.
Can I run a local AI on 8 GB of RAM?
Yes. Stick to models around 4B parameters, like qwen3.5:4b or gemma3:4b, and close memory-hungry apps first.
Which is better, Qwen or Gemma?
Both are strong at these sizes. Try one for a day, then the other. It takes one command to switch, and you'll quickly find which style of answer you prefer.
