RaMem: Contextual Reinstatement for Long-term Agentic Memory

2026-06-22

AI Summary

This work addresses the problem of “context collapse” in large language model agents during long-term interactions, where retrieved memory fragments lack their original contextual grounding, rendering them relevant yet unverifiable for current validity. To mitigate this, the authors propose RaMem, a novel framework that introduces a context reconstruction mechanism. RaMem integrates multidimensional contextual cues—such as temporal order, conversational span, and participant roles—through four stages: evidence anchoring, recall-conditioned inference, validity-aware retrieval, and context-preserving synthesis, thereby transforming raw memories into verifiable contextual evidence. Experimental results demonstrate that RaMem achieves an average F1 improvement of over 10% against strong baselines across multiple long-term memory benchmarks, significantly enhancing the agent’s memory reliability and reasoning capabilities.

Abstract

Long-term memory has become increasingly important for LLM agents that operate across extended interactions and evolving task contexts. Recent memory systems have made past experiences more persistent, compact, and retrievable, but retrieval alone does not ensure that a memory provides valid evidence for the current query. When experiences are compressed into reusable fragments, memories from different situations may appear equally relevant if they involve recurring entities or user states. We refer to this failure as context collapse: memories lose the surrounding context needed to judge whether they provide valid evidence for the current query. To address this problem, we propose Contextual Reinstatement for Agentic Memory (RaMem), a framework that turns retrieved memory fragments into contextually verifiable evidence. RaMem operates through four coordinated stages: (i) evidence anchoring grounds each memory in its original episodic conditions, especially event time, mention time, session span, and participants; (ii) recall condition induction derives the evidence conditions implied by the query; (iii) validity-aware retrieval uses these conditions to prioritize context-compatible memories while retaining content-relevant candidates as fallback evidence; and (iv) context-preserved synthesis keeps the selected memories' structured context available to the generator. Experiments on long-term memory benchmarks show that RaMem consistently improves performance over strong memory baselines, with average F1 gains of more than 10% across several backbones.

Problem

• long-term memory
• context collapse
• LLM agents
• memory retrieval
• contextual validity

Innovation

• contextual reinstatement
• long-term agentic memory
• context collapse
• validity-aware retrieval
• evidence anchoring