Ollama
FreeRun large language models on your own machine, one command away
An MIT-licensed local LLM runner. It downloads models, manages VRAM and exposes a local API that has become the de-facto standard interface — most local AI tools support Ollama first. No GUI: you get a command and a local service. No account, works offline.
Why we picked it
Ollama is the default answer for running models locally. 178,502 stars, MIT-licensed, the widest model library, and whatever tool you want to plug in probably supports it already.
One number to know first: the Windows installer is 1492 MB, versus 174.6 MB on macOS. The gap is CUDA and ROCm GPU runtimes bundled into the Windows build. Leave room before installing — and that is before any model, where a 7B typically adds another 4-5 GB.
There is no GUI. If you want a window you can just click and chat in, pair it with a front end like Open WebUI, or use LM Studio or Jan instead. But the moment you want a local model wired into an editor, a script or your own program, Ollama's local API is the path of least resistance.