EvoGraph-Mem: Failure-Aware Editable Graph Memory for Long-Term Language Agents

2026-08-03

AI Summary

This work addresses the vulnerability of long-term language agents to memory contamination during sustained interactions, which can lead to outdated, overgeneralized, or harmful stored insights that degrade performance. To mitigate this, the authors propose an editable insight graph memory system that dynamically maintains and retrieves knowledge by explicitly tracking positive and negative evidence for each insight along with its activation status. The system incorporates a fault-aware memory maintenance mechanism that enables retention, archival, revision, and addition of memories—overcoming the limitations of traditional append-only memory architectures. It further integrates graph-structured representation, utility-aware retrieval, and evidence-driven update strategies. Experiments demonstrate consistent and significant improvements over existing memory-augmented baselines across multiple backbone models, with ablation studies confirming the effectiveness of graph-level editing and evidence-aware retrieval in enhancing both memory reliability and task performance.

Abstract

Long-term memory is essential for language agents operating across extended interactions and evolving tasks. Existing memory-augmented agents mainly focus on storing and retrieving past experience, but the quality of stored memories may degrade over time. In particular, previously distilled insights can become outdated, over-generalized, or harmful under new task contexts, causing memory pollution when repeatedly reused. To address this issue, we study insight-level memory maintenance for long-term language agents and propose a failure-aware memory maintenance framework based on an editable insight graph. Each insight node tracks positive evidence, negative evidence, and an activation state, enabling the agent to distinguish reusable insights from conflicting or invalid ones. We further introduce a utility-aware retrieval mechanism and a graph controller that updates the memory graph after task execution by keeping reliable insights, archiving invalid ones, revising outdated ones, and adding newly discovered reusable insights. Extensive experiments show that our method consistently outperforms representative memory-based agent baselines across different backbone models. Ablation studies further demonstrate that append-only memory is insufficient for long-horizon tasks, while evidence-aware retrieval and graph-level editing improve memory reliability and downstream task performance.

Problem

• long-term memory
• memory pollution
• insight maintenance
• language agents
• memory degradation

Innovation

• editable graph memory
• failure-aware memory
• insight-level maintenance
• evidence-aware retrieval
• memory pollution