I've been meaning to run a local model on my Mac for a while but always bounced off the setup - picking the right quantization, figuring out what my hardware could actually handle, getting a decent frontend on top of it. Local just skips all of that. It looked at my machine, picked models that actually run well, and I had a working local chat in a few minutes with zero config decisions to make. The speed difference versus the same models run through a generic wrapper is very noticeable too, meeting notes transcribe close to real time on my M-series Mac.