It works fine with local models. I have it pointing at my strix halo running Qwen3.8 Flash Next, for example.