Nous: A Predictive World Model for Long-Term Agent Memory

2026-06-20

AI Summary

This work addresses the limitations of traditional agent memory systems, which rely on static storage and struggle to support dynamic knowledge evolution and forgetting in long-term dialogue. The authors propose a memory architecture grounded in a predictive world model, representing knowledge as probability distributions over entity–attribute pairs. Rather than storing factual statements, the system tracks belief evolution through Bayesian updates and naturally implements forgetting via information entropy. The approach eschews explicit storage, instead incorporating mutual information–driven entity disambiguation and entropy-based decay mechanisms, eliminating the need for external vector databases or graph engines. Evaluated on the LoCoMo benchmark using GPT-4o-mini, the method achieves F1 scores of 63.50 (single-hop), 55.32 (multi-hop), 58.57 (temporal), and 62.50 (open-domain), outperforming A-MEM and BeliefMem on most metrics.

Abstract

We present Nous, a novel agent memory architecture grounded in the principle that knowledge is prediction, not storage. Rather than persisting facts as database records, vector embeddings, or knowledge-graph triples, Nous maintains a predictive world model: a collection of categorical probability distributions, called dimensions, one per entity-attribute pair observed in conversation. Each incoming observation is scored by its information-theoretic surprise S = -log2 P(obs | D), and the distribution is updated via a closed-form Bayesian posterior. The primary stored artifact is the delta, a record of the shift from prior to posterior belief, rather than the fact itself. Forgetting emerges naturally as entropy decay toward the uniform distribution, and identity resolution is handled through mutual information between entity dimension sets. Evaluated on the LoCoMo long-term conversational memory benchmark across ten conversations (1,540 questions) using GPT-4o-mini as backbone, Nous achieves F1 of 63.50 (single-hop), 55.32 (multi-hop), 58.57 (temporal), and 62.50 (open-domain). Against A-MEM's self-reported GPT-4o-mini numbers, Nous shows substantial gains in three of four categories, though we note that independent citations of A-MEM's results disagree with each other on category assignment, a reproducibility issue we discuss openly rather than resolve unilaterally. We additionally compare against BeliefMem, a concurrently developed system built on the same core premise of belief-based rather than deterministic memory; on the same benchmark and backbone, Nous's self-reported numbers exceed BeliefMem's self-reported numbers on all four categories, though we flag several uncontrolled differences between the two evaluation pipelines that prevent this from being a fully controlled comparison. Nous requires no external vector database or graph engine.

Problem

• agent memory
• predictive world model
• long-term memory
• conversational AI
• memory architecture

Innovation

• predictive world model
• belief-based memory
• Bayesian updating
• information-theoretic surprise
• entropy-based forgetting