The five steps at a glance
1. Find out what you have
Open My hardware and press Detect. The number that matters is memory: video memory on a graphics card, unified memory on a Mac, RAM on a phone or board.
2. Pick a model that fits
The Model Finder lists every model that runs on your hardware, best first. Choose one that says Runs great or Runs well.
3. Install an app
LM Studio is point and click. Ollama is one command. Each model page names the best app for your exact hardware.
4. Download and chat
Use the “Open in” button for your app, or copy the model name into the app’s search box. Start with the quantization the page recommends.
5. Check the speed
If answers feel slow, step down one quantization or choose a smaller model. If you have headroom, step up.
- Find out what you have. Open My hardware and press Detect. The number that matters is memory: video memory on a graphics card, unified memory on a Mac, RAM on a phone or board.
- Pick a model that fits. The Model Finder lists every model that runs on your hardware, best first. As a rule, choose one with a quality grade of A or better that says Runs great or Runs well. Not sure how much memory a model needs? Try the memory calculator.
- Install an app. LM Studio is point and click, Ollama is one command. Each model page names the best app for your exact hardware. See all apps. On a phone or tablet, install PocketPal AI from the App Store or Google Play instead.
- Download and chat. On a model page, use the “Open in” button for your app, or copy the model name into the app’s search box. Start with the quantization the page recommends. On a phone, tablet or other mobile device there is nothing to copy: the model page lists the taps and the one file to download inside the app (mobile guide).
- Check the speed. If answers feel slow, step down one quantization or choose a smaller model. If you have headroom, step up: quantization explained.
Everything runs on your machine. No account is needed, and your prompts stay local.