Trail brings local smarts to smart homes

🔥 What happened Researchers at UIUC unveiled Trail, a fully local framework for smart home control. Instead of open-ended generation, it makes hierarchical typed decisions: room, device, operation – pulled straight from the runtime device registry. 💡 Why it matters Trail hits 85.3% accuracy on altered home descriptions, versus 69.8% for a finetuned Qwen3.5-4B baseline. With memory, it resolves 75.6% of follow-up requests; the baseline manages just 42.5%. Structure beats model size for local voice control. ⚡ Our take Anyone still routing smart-home AI through the cloud either has no privacy budget or too much faith in open-ended generation. Trail proves typed decisions are the smarter path.