Voice and chat assistants face the complex task of accurately converting messy natural language into precise, structured commands. Domux is an open model developed by iFlytek specifically for smart-home command understanding. Its core function involves intent parsing and slot filling to interpret user requests like "Make it cozy in here." The model is fine-tuned on google/gemma-4-E2B-it, making it a compact base model. Domux supports multimodal input, processing both images and text. Its primary target deployment is on-device or at the edge, rather than relying on large cloud models. This focus on compact size is crucial for reducing round-trips and enabling more localized interaction within the home. Keeping command understanding on the device enhances efficiency and user experience. Domux is accessible via its model card on Hugging Face, though access may be gated. This open work highlights the importance of practical, task-focused on-device models in the broader AI landscape. It demonstrates that not every AI solution requires a giant cloud-based model for effective operation.