Running your first model locally

You do not need to be a programmer to run a model on your own machine. Two free tools have made this genuinely easy, and the whole exercise takes about fifteen minutes including the download.

LM Studio if you want an app. It looks like a chat app: browse models inside it, click download, start typing. It shows you which models fit your machine before you download, and everything happens in a window. This is the right first tool if the command line is not your happy place.

Ollama if you want a tool. One install, then ollama run followed by a model name in a terminal. It downloads the model the first time and starts a chat. Its real strength is what comes next: it quietly serves an API on your machine, so anything that can talk to a model — editors, scripts, coding agents — can talk to yours. This is the right tool if you plan to build anything.

Both pull the same underlying files and both run the same open-weight models, so the choice is temperament, not capability.

Which model to start with. Something mid-sized and well-rated that fits your memory — the catalogue's fit verdicts exist for exactly this question, and every model page shows what your machine can hold at the standard quantization. A rule of thumb while you learn: a model around 4B runs on almost anything made this decade, one around 8B wants a decent laptop, and 20B-class models want a real graphics card or an Apple Silicon Mac with plenty of memory.

What to expect. The first response feels slow compared with a hosted service, and that is mostly your hardware being honest with you — the speed estimates on our fit pages tell you beforehand. What you get in exchange: nothing you type leaves the machine, ever, and nobody meters you.

Where next: Check what your machine runs · Point your coding tools at it