SAGE: A Novelty Gate for Efficient Memory Evolution in Agentic LLMs

2026-05-28

AI Summary

This work addresses the challenge of inefficient write control in long-term memory for large language model (LLM) agents by framing memory evolution as a novelty detection task. The authors propose a novelty-aware writing mechanism based on the von Mises–Fisher distribution, which operates in a spherical embedding space. Their approach employs adaptive gating to score candidate facts and dynamically sets thresholds according to the geometric structure of the memory bank, invoking the LLM for memory consolidation only under high uncertainty. Experimental results demonstrate that the method outperforms Mem0 on the LoCoMo benchmark, reducing API costs by 3.4× and latency by 2.5× when using GPT-4o-mini. Furthermore, integrated as an A-Mem plugin, it eliminates 16–18% of LLM calls while preserving memory quality with negligible degradation.

Abstract

Agentic LLMs must continuously decide whether newly extracted facts should be added, merged with existing memories, or ignored, yet prior work has focused more on retrieval and storage than on principled write-side control. We frame memory evolution as a novelty-detection problem and propose SAGE, a Spherical Adaptive Gate for memory Evolution that scores candidate facts with a von Mises-Fisher-based density estimator over memory embeddings and routes them with an adaptive threshold that tracks memory-store geometry. SAGE resolves clearly novel facts as ADD, clearly redundant facts as NOOP, and sends only uncertain cases to an LLM merge step, reducing expensive write-time reasoning. On LoCoMo, SAGE achieves the best average token-F1 against Mem0 on all seven open-weight backbone comparisons, while on GPT-4o-mini it reduces add-phase API cost by 3.4$\times$ and add-phase latency by 2.5$\times$ with only a small average judge-score gap. As a drop-in binary gate for A-Mem, SAGE skips roughly 16-18% of LLM calls across five models with minimal quality change on open-weight backbones. These results suggest that novelty-aware write control is a practical lever for improving both memory quality and system efficiency in long-term agentic memory.

Problem

• Agentic LLMs
• memory evolution
• novelty detection
• write-side control
• memory management

Innovation

• novelty detection
• memory evolution
• adaptive threshold
• von Mises-Fisher
• agentic LLMs