Beyond Static Summarization: Proactive Memory Extraction for LLM Agents

TL;DR

This paper proposes proactive memory extraction (namely ProMem), which introduces a recurrent feedback loop where the agent uses self-questioning to actively probe the dialogue history and achieves a superior trade-off between extraction quality and token cost.

Abstract

Memory management is vital for LLM agents to handle long-term interaction and personalization. Most research focuses on how to organize and use memory summary, but often overlooks the initial memory extraction stage. In this paper, we argue that existing summary-based methods have two major limitations based on the recurrent processing theory. First, summarization is "ahead-of-time", acting as a blind "feed-forward" process that misses important details because it doesn't know future tasks. Second, extraction is usually "one-off", lacking a feedback loop to verify facts, which leads to the accumulation of information loss. To address these issues, we propose proactive memory extraction (namely ProMem). Unlike static summarization, ProMem treats extraction as an iterative cognitive process. We introduce a recurrent feedback loop where the agent uses self-questioning to actively probe the dialogue history. This mechanism allows the agent to recover missing information and correct errors. Our ProMem significantly improves the completeness of the extracted memory and QA accuracy. It also achieves a superior trade-off between extraction quality and token cost.

Figures (5)

• Figure 1: General framework of summary-based memory management for LLM agents. It starts with memory extraction, where raw dialogues are compressed into summaries. This is followed by memory organization and utilization, which includes storage, evolutionary updates, and retrieval. While most research focuses on the latter stage, the extraction phase is often overlooked and treated as a simple, one-way preprocessing step.
• Figure 2: The connection between RPT and our method. (a) The Kanizsa illusion shows how the brain uses recurrent processing to form a complete shape from parts, adapted from ConsciousnessInAI. (b) The high-level idea of our proactive memory extraction. Our method uses a feedback loop to construct a complete and verified memory summary from raw dialogue history.
• Figure 3: The overview of our proposed proactive memory extraction framework. The workflow consists of three main stages: (1) Initial Extraction, where the agent performs basic fact extraction from dialogue history; (2) Memory Completion, which uses similarity matching to align extracted entries with original context; and (3) Recurrent Verification, where the agent proactively generates self-questions to probe the raw dialogue for supplementary information. Finally, a deduplication step ensures a high-quality and concise final memory.
• Figure 4: Performance w.r.t. compression ratios.
• Figure 5: Performance comparison on LongMemEval.