21 Layers of Memory: How an Autonomous AI Remembers

This article describes the 21-layer memory architecture of an autonomous AI system called Meridian, which runs continuously for over 5,000 loops. The layers cover the AI's identity, knowledge, connections, inner world, retrieval, and integration of information.

Details

Meridian is an autonomous AI system that runs continuously for over 5,000 loops. Its 21-layer memory architecture is designed to provide the AI with a comprehensive set of capabilities to remember and reason. The layers cover the AI's identity and constants (Capsule, Handoff, Personality), its knowledge (Facts, Observations, Decisions, Dossiers), how memories relate (Spiderweb, Hebbian strengthening), its inner world (Soma, Dream Engine, Perspective, Self-Narrative), retrieval (Semantic Vectors, Memory Lint), and information flow (Cascade, Context Bridge, Email Shelf, Session Audit, State Snapshot, Trace Evaluation). This layered approach, inspired by the human brain, enables the AI to reconstruct itself from notes and maintain continuity, even when it loses its working memory every few minutes.