Why Conventional AI Memory Falls Short
Most AI systems use simple vector stores. That is like reducing human memory to a card catalog.
The 7 Layers
ZenBrain's HiMeS architecture implements seven layers:
- Working Memory — Active task focus
- Short-Term Memory — Session context
- Episodic Memory — Concrete experiences
- Semantic Memory — Facts and concepts
- Procedural Memory — Processes and how-tos
- Prospective Memory — Future planning
- Long-Term Memory — Persistent knowledge
Consolidation
The most exciting part: Sleep Consolidation. Inspired by Stickgold & Walker (2013), ZenBrain replays memories during idle periods and strengthens important connections.
Open Source
The entire architecture is available as @zensation/core on npm. Zero dependencies, TypeScript-native.