AdaHome: An Adaptive Smart Home Assistant using Local Small Language Models

2026-07-20

AI Summary

This work addresses the limitations of existing large language model–based smart home assistants, which rely on cloud deployment and complex reasoning, rendering them inefficient on resource-constrained devices and lacking continuous personalization. The authors propose an adaptive assistant framework tailored for local small language models, featuring an intent-aware routing mechanism that dynamically selects between lightweight reasoning and direct prompting strategies. Additionally, a Chain-of-Draft decision-making mechanism enables efficient task planning. Notably, the approach achieves preference adaptation solely through user feedback, without requiring prompt engineering or model retraining. Experimental results demonstrate that the system attains 86.7% accuracy on direct command tasks with latency reduced to one-third of the baseline, and achieves 88% preference consistency in multi-turn interactions—substantially outperforming a prompt-augmented baseline (52.5%).

Abstract

Smart home assistants interpret a wide range of user commands, from explicit device control to underspecified and preference dependent requests. While recent systems based on Large Language Models (LLMs) improve this capability, they often rely on heavyweight reasoning pipelines and cloud-based deployment, limiting their efficiency and suitability for resource-constrained environments, and raising privacy concerns. In addition, existing approaches provide limited support for stable long-term personalization. To address these issues, we present AdaHome, an adaptive smart home assistant designed for locally deployed small language models in smart home environments. Rather than applying complex reasoning uniformly, AdaHome introduces an intent-aware planning framework that dynamically routes commands either to straightforward prompt-based or lightweight reasoning-based components. For commands requiring interpretation, we adopt a Chain-of-Draft strategy to enable efficient and stable decision-making. To support personalization, we further propose a preference adaptation mechanism that learns from user feedback over time without requiring prompt augmentation or model retraining. We evaluate AdaHome against representative LLM-based baselines under a unified small model setting. AdaHome achieves substantially higher accuracy on direct commands (86.7%) while reducing latency by up to 3$\times$. Furthermore, it maintains competitive performance on ambiguous inputs with lower computational cost. In multi-turn scenarios, AdaHome achieves 88% preference consistency, compared to 52.5% for a prompt augmentation baseline.

Problem

• smart home assistant
• local deployment
• personalization
• resource-constrained environments
• privacy concerns

Innovation

• local small language models
• intent-aware planning
• Chain-of-Draft
• preference adaptation
• on-device personalization