Core editing runs locally. Built-in Local AI runs locally after model download. If you choose OpenAI or another configured endpoint, relevant prompt and diagram context are sent to that provider.