I suggest adding user-configurable hardware selection for the local AI email writing assistant. Currently, it automatically chooses the GPU, limiting optimization. My system includes an NPU, an NVIDIA GPU, and an AMD GPU. I’d like to set the NPU as default for lightweight, energy-efficient writing, especially on battery. For longer or more complex drafts, users should be able to select a preferred GPU or fall back to NVIDIA/AMD when available or high demand scenaries. This would improve efficiency, control, and performance.