Personal Health Agent (PHA; Google multi-agent framework over wearable + medical data)
- Year
- 2026
- Form factor
- Wrist
Activity counts measure attention, not model quality. Missing counts are unknown. Publication status follows the linked source; a listed venue alone does not confirm peer review.
Personal Health Agent is Google's open multi-agent framework for personal health questions over consumer wearable and medical context. An orchestrator coordinates data-science, domain-expert, and health-coach LLM sub-agents with iterative reflection. Apache-2.0 code is available; the paper runs used Gemini. Published in Nature 2026 (DOI 10.1038/s41586-026-11161-8).
- Form factor
- Wrist
- Year
- 2026
- Architecture
Orchestrator coordinating three LLM sub-agents (Data Science agent with sandboxed code execution over personal time series; Domain Expert agent with web search + Data Commons; Health Coach agent for multi-turn goal setting) with iterative reflection; baselines in the repo: Parallel (no orchestration) and PHIA (single ReAct agent). Model-agnostic; Gemini in the paper
- Signals
EHR, Text, activity records, wearable time series (consumer wellness devices)
- Pretraining data
n/a. Evaluated on WEAR-ME (n=1,165; 50-persona subset open for research use in the repo) and an n=2,000 symptom dataset
- License
Apache-2.0 (personal_health_agent/LICENSE.txt). No weights: model-agnostic framework; the paper's runs used Gemini (closed).
- Type
- Task model
- Form factor
- Wrist
- Parameters
- Not reported
- Architecture
- Orchestrator coordinating three LLM sub-agents (Data Science agent with sandboxed code execution over personal time series; Domain Expert agent with web search + Data Commons; Health Coach agent for multi-turn goal setting) with iterative reflection; baselines in the repo: Parallel (no orchestration) and PHIA (single ReAct agent). Model-agnostic; Gemini in the paper
- Venue
- Nature, 2026-10-21 (preprint arXiv 2508.20148)
- Year
- 2026
- Weights
- Not released
- Licence
- Apache-2.0 (personal_health_agent/LICENSE.txt). No weights: model-agnostic framework; the paper's runs used Gemini (closed).
- Caveats
- Not reported
- Datasets
- Not reported
- Added
- Oct 22, 2026
- Updated
- Oct 26, 2026
Abstract
Health is a fundamental pillar of human wellness, and the rapid advancements in large language models (LLMs) have driven the development of a new generation of health agents. However, the application of health agents to fulfill the diverse needs of individuals in daily non-clinical settings is underexplored. In this work, we aim to build a comprehensive personal health agent that is able to reason about multimodal data from everyday consumer wellness devices and common personal health records, and provide personalized health recommendations. To understand end-users' needs when interacting with such an assistant, we conducted an in-depth analysis of web search and health forum queries, alongside qualitative insights from users and health experts gathered through a user-centered design process. Based on these findings, we identified three major categories of consumer health needs, each of which is supported by a specialist sub-agent: (1) a data science agent that analyzes personal time-series wearable and health record data, (2) a health domain expert agent that integrates users' health and contextual data to generate accurate, personalized insights, and (3) a health coach agent that synthesizes data insights, guiding users using a specified psychological strategy and tracking users' progress. Furthermore, we propose and develop the Personal Health Agent (PHA), a multi-agent framework that enables dynamic, personalized interactions to address individual health needs. To evaluate each sub-agent and the multi-agent system, we conducted automated and human evaluations across 10 benchmark tasks, involving more than 7,000 annotations and 1,100 hours of effort from health experts and end-users. Our work represents the most comprehensive evaluation of a health agent to date and establishes a strong foundation towards the futuristic vision of a personal health agent accessible to everyone.
Verbatim fromarXiv· arXiv metadata (abstracts are released under CC0)
How it works
- Pretraining data
- n/a. Evaluated on WEAR-ME (n=1,165; 50-persona subset open for research use in the repo) and an n=2,000 symptom dataset
- Architecture
- Orchestrator coordinating three LLM sub-agents (Data Science agent with sandboxed code execution over personal time series; Domain Expert agent with web search + Data Commons; Health Coach agent for multi-turn goal setting) with iterative reflection; baselines in the repo: Parallel (no orchestration) and PHIA (single ReAct agent). Model-agnostic; Gemini in the paper
- Inputs
- EHR, Text, activity records, wearable time series (consumer wellness devices)
- Approach
- LLM agent · multi agent · personal health · health coach · data science agent · wearables
Datasets
Cite
APA
Heydari, A. A., Gu, K., Srinivas, V., Yu, H., Paruchuri, A., Zhang, Z., Zhang, Y., He, Q., Palangi, H., Hammerquist, N., Metwally, A. A., Winslow, B., Kim, Y., Ayush, K., Yang, Y., Narayanswamy, G., Xu, M. A., Garrison, J., Schmidgall, S., . . . Xu, X. O. (2026). The anatomy of a personal health agent. Nature, 2026-10-21. https://doi.org/10.1038/s41586-026-11161-8
BibTeX
@article{heydari2026anatomy,
title = {{The anatomy of a personal health agent}},
author = {Heydari, A. Ali and Gu, Ken and Srinivas, Vidya and Yu, Hong and Paruchuri, Akshay and Zhang, Zhihan and Zhang, Yuwei and He, Qian and Palangi, Hamid and Hammerquist, Nova and Metwally, Ahmed A. and Winslow, Brent and Kim, Yubin and Ayush, Kumar and Yang, Yuzhe and Narayanswamy, Girish and Xu, Maxwell A. and Garrison, Jake and Schmidgall, Samuel and Lee, Amy Armento and Vafeiadou, Jenny and Graef, Ben and Galatzer-Levy, Isaac R. and Schenck, Erik and Barakat, Andrew and Perez, Javier and Shreibati, Jacqueline and Hernandez, John and Faranesh, Anthony Z. and Rezaei, Naghmeh and Prieto, Javier L. and Heneghan, Conor and Zhan, Jiening and Malhotra, Mark and Patel, Shwetak and Liu, Yun and Althoff, Tim and Liu, Xin and McDuff, Daniel and Xu, Xuhai Orson},
year = {2026},
journal = {Nature, 2026-10-21},
doi = {10.1038/s41586-026-11161-8},
eprint = {2508.20148},
archivePrefix = {arXiv}
}Built from the hub record. Check it against the publisher before submitting.
History
Record updated
Added to the ChatPPG research hub
Connections
Open in the graph explorer
The anatomy of a personal health agentShared topic: personal healthNature, published 2026-10-21 · 2026
Agentic AI-enabled discovery across large-scale sleep physiology (AI Sleep Co-Scientist)Shared topic: multi agentarXiv preprint · 2026
An AI Co-Data-Scientist for Prioritizing Candidate Biomarkers from Wearable Sensor Data (CoDaS)Shared topic: multi agentarXiv 2604.14615 · 2026- AutoResearcHAR: A Multi-Agent Framework for Dataset Adaptive IMU Based Human Activity RecognitionShared topic: multi agentISWC '26 · 2026