Local-First Smart Home Applications with HubOS

Many smart home applications collect sensitive data and interact with remote services. The local-first principle minimizes data propagation by favoring local processing and communicating with remote services only when necessary. HubOS is an operating system for smart home hubs supporting developers in the design and execution of privacy-conscious, local-first applications. HubOS applications can combine smart home computations and automation modules of diverse complexity, from simple trigger-action rules to CPU-intensive operations such as inference from machine-learned AI models. For the latter, it leverages WebAssembly, enabling portability and near-native performance. All HubOS application modules are strictly isolated and controlled in their access to resources and data. We contribute a novel Trigger-Action-Based Access Control model (TABAC) that allows context-sensitive and user-understandable control of access conditions. We demonstrate the interest of HubOS by detailing the implementation of three use case applications: a voice assistant, a security alarm based on face recognition, and a fall detector with remote video assistance for elderlies. Our evaluation shows that even for smart hubs with modest resources, HubOS applications have sufficient performance to contend with cloud-only alternatives.