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

2026-07-09

AI Summary

This work addresses the challenge in long-horizon tasks where decision-relevant information tends to decay or exceed context window limits as trajectories grow, thereby degrading agent performance. To mitigate this, the authors propose an active memory agent operating in parallel with the action agent, which stores critical information in a structured memory bank and proactively intervenes in the decision-making process through selective and timely memory injection, rather than passive retrieval. The approach integrates seamlessly as a plug-and-play module and is trained on Qwen3.5-27B using supervised fine-tuning (SFT) and GRPO. Experiments demonstrate significant improvements, achieving +8.3 and +6.8 percentage points in pass@1 on Terminal-Bench 2.0 and τ²-Bench, respectively, along with partial cross-benchmark transferability.

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.

Problem

• long-horizon tasks
• behavioral state decay
• decision-relevant state
• memory management
• context window

Innovation

• proactive memory agent
• behavioral state decay
• long-horizon tasks
• structured memory bank
• selective intervention