Quantization should not be exclusive to any particular inference server. oQ produces standard mlx-lm compatible models that work everywhere — oMLX, mlx-lm, and any app that supports MLX safetensors.