Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents

Yifan Wu, Lizhu Zhang, Yuhang Zhou, Mingyi Wang, Bo Peng, Serena Li, Xiangjun Fan, Zhuokai Zhao

arXiv:2607.08716 · 2026-07-11 공개 · arXiv · PDF

grpo sft terminal-bench tau2-bench long-horizon-agents memory-agent proactive-memory behavioral-state-decay

Abstract

In long-horizon tasks, decision-relevant state is often scattered across an expanding trajectory, while the action agent must surface it and act. As trajectories grow, task requirements, environment facts, prior attempts, diagnoses, and open subgoals can be buried in the context window or pushed beyond it, failing to influence decisions when needed. We call this failure mode "behavioral state decay". We study memory as an active intervention mechanism rather than passive retrieval. A separate memory agent runs alongside an unmodified action agent, updating a structured memory bank from the recent trajectory and deciding whether to inject a memory-grounded reminder or remain silent. The module is plug-and-play with frontier action agents and existing agent harnesses. Across Terminal-Bench 2.0 and τ^2-Bench, it improves pass@1 for both weaker and stronger action agents, with gains of +8.3 pp on Terminal-Bench and +6.8 pp on τ^2-Bench. Ablations show that selective intervention outperforms passive bank exposure, always-on injection, advisor-only guidance, and general retrieval. As an early step toward open-weight memory policies, we train Qwen3.5-27B on SETA using SFT and GRPO, improving validation reward and achieving partial transfer to Terminal-Bench.

한국어 요약

한 줄 요약

긴 시간 범위의 작업에서 행동 에이전트와 별도로 작동하는 메모리 에이전트가 행동 성능을 +8.3pp 향상시킨다.

핵심 기여도

핵심 아이디어

기존 메모리 시스템은 정보 저장 및 검색에 집중하지만, 장기적 작업에서는 **언제** 메모리를 사용할지 결정하는 것이 핵심이다. 이 논문은 메모리를 **선택적 개입**(intervention)으로 정의하고, **메모리 에이전트**(memory agent)를 도입하여 **실행 상태**(execution state)를 유지하고, 필요 시 **구체적 리마인더**(예: 실패 진단, 미완료 서브목표)를 주입한다. 이는 단순히 메모리를 노출시키는 **Full-bank context**나, 항상 주입하는 **Always inject** 방식보다 **선택적 침묵**(silence)을 포함한 정책이 더 효과적임을 보여준다.

기술적 접근법

주요 결과

의의 및 한계

실용적 활용