yg1112

The key difference is that MLX's array model assumes unified memory from the ground up. llama.cpp's Metal backend works fine but carries abstractions from the discrete GPU world — explicit buffer synchronization, command buffer boundaries — that are unnecessary when CPU and GPU share the same address space. You'll notice the gap most at large context lengths where KV cache pressure is highest.

From the repo, it signs natively on macOS and falls back to a cloud signing server (shortcut-signing-server). That fallback matters -- without macOS you would have to reverse-engineer Apple signing format yourself, and it changes across iOS versions. The hosted signing server is really what makes the whole cross-platform toolchain viable.