User-Controllable Hierarchical Memory System for Personal AI Assistants

Current AI assistants and large language models suffer from two critical limitations: short context window and lack of user control over memory. These issues lead to poor long-term user experience and serious privacy concerns. This paper proposes a User-Controllable Hierarchical Memory System (UCHMS) designed specifically for personal AI assistants. The system divides memory into three layers: Temporary Memory, Optional Memory, and Long-term Core Memory. Unlike existing memory mechanisms that focus on improving AI agent performance, this system is user-centric, giving users full authority to manage their own data. This design improves personalization, privacy security, and long-term interaction experience.