Researchers have built an AI agent that monitors continuous heart and activity data from wearables — and can both answer questions about it and raise alerts without being prompted.
The system, called VitalAgent, is a tool-augmented framework designed to work with ECG and PPG signals from wearable devices. Unlike most mobile health apps, which either run a fixed prediction model or let users ask questions about a static data snapshot, VitalAgent maintains a running physiological memory and can execute dynamic computations over raw signal streams. The team also released VitalBench, a benchmark comprising 1,862 question-answer pairs and 90.2 hours of continuous ECG and PPG recordings, covering cardiac health, physical activity, and stress. Against prompt-based and ReAct baselines, VitalAgent posted more than a 25% improvement on reactive evaluation tasks.
The gap this targets is real: most consumer health wearables surface daily summaries and push the interpretation onto the user. An agent that can hold context across weeks of physiological data and escalate proactively is a meaningfully different capability — closer to a background clinician than a fitness dashboard. Whether that framing survives contact with regulatory scrutiny is another question entirely.
Health AI has a long track record of impressive benchmark numbers that do not translate to clinical utility, and wearable signal quality is notoriously variable outside controlled settings — so treat the 25% gain as a research result, not a product promise.