To improve long-term user-AI collaboration, I propose a tiered memory structure:
Primary Memory (Cloud-Based)
Persistent and synced across devices
Stores core facts and context (already exists in limited form)
Secondary Memory (Local or Device-Specific)
Cached memory stored locally (e.g., in the app or browser)
Larger capacity (even just 10MB of text goes a long way)
Activated only on designated device(s) for security
Sync optional—designed for context-rich continuity without overloading the cloud system
Key Benefits:
Enables deeper, evolving interactions without constant repetition
Users can promote/demote memories between primary and secondary
Offers personalization while respecting compute limits and privacy
Scales naturally across free and paid tiers (e.g., no secondary for free, 10MB+ for paid)
This hybrid system would empower serious users (like writers, researchers, neurodiverse users, disabled users, etc.) to form long-term, adaptive relationships with their AI without compromising speed, simplicity, or safety.