Memory Systems

Persistent, evolving memory for long-lived AI

From Ebbinghaus-inspired forgetting curves to LLM-maintained wikis — how agents remember, forget, and compound knowledge over time.

LLM as OS, virtual context management, autonomous memory paging.

Ebbinghaus forgetting curve for LLM memory, important memories reinforced, old ones fade.

Long-term memory for AI self-evolution, OMNE #1 on GAIA benchmark (40.53% vs GPT-4's 15%).

Search→Synthesize→Evolve cycle, ReMem pipeline, 92% on BabyAI.

LLM-maintained personal wiki as a RAG alternative — write-side synthesis, persistent compounding artifact.