HealthClaw: Self-Evolving AI for Personal Health.

Researchers have developed HealthClaw, an innovative open-source agent architecture specifically designed for long-term personal health management. Unlike most health AI systems that treat each user request in isolation, HealthClaw is built to adapt and evolve its support as a person's routines, preferences, measurements, and health risks change over time. The architecture intelligently separates shared safety rules and medical knowledge from a private, longitudinal memory that stores individual profile facts, reusable procedures, and episodic traces. After each interaction, an induction process determines what information should update the user's profile, revise existing procedures, remain as episodic memory, or be excluded entirely. Evaluations using a synthetic year-long benchmark and multiple biomedical tasks demonstrated significant improvements. HealthClaw increased answer accuracy from 0.2% to 45.7% compared to current-query prompting, while reducing context exposure by 71.7% compared to full-history prompting. It also showed superior privacy-aware answer quality and fewer unsafe disclosures, supporting the concept of governed, self-evolving memory for personal health agents.