@Tadano >github link
Software on github often turns out to be proprietary.

The license is apparently MIT expat, which expressively gives permission to sublicense and therefore permission to make sections proprietary, which usually ends up happening.

As typical of proprietary software, the install instructions are literally; "curl -fsSL https://ollama.com/install.sh | sh", which goes and installs proprietary CUDA libraries if a nvidia GPU is detected and seemingly proprietary AMD ROCm libraries if a AMD GPU is detected.

It at least supports a CPU-only mode, but then the question is the model and the vast majority of trained models infringe copyright, which makes those models proprietary.