Running models locally — LM Studio, Ollama, Jetson-class hardware, the broader case for keeping inference on hardware you own.