Learning What to Remember and What to Internalize in LLM Self-Evolution via Adaptive Memory-Parameter Coordination

2026-08-02

AI Summary

This work addresses the challenge that large language models struggle to simultaneously maintain flexibility and performance in dynamic environments, where continuous changes in tool interfaces, APIs, and user requirements undermine static memory or parameter-update mechanisms. To overcome this limitation, the authors propose COVE, a novel framework that synergistically integrates memory-based and parameter-based self-evolution strategies for the first time. COVE dynamically allocates learning pathways through task-aware routing, phase-aware scheduling, and knowledge refinement, effectively combining editable external memory with model fine-tuning. Experimental results demonstrate that COVE significantly outperforms single-path approaches across diverse dynamic tasks, achieving more robust and efficient capability evolution.

Abstract

Large language model agents increasingly operate in dynamic environments where tool interfaces, APIs, and user requirements change after deployment. Existing self-evolution methods mainly follow two paradigms: harness-based approaches, which externalize feedback into editable memories or skills for rapid adaptation, and parameter-based approaches, which internalize experience into model parameters for deeper capability improvement. However, using either mechanism alone creates a trade-off between flexibility and performance. This paper asks how an agent can coordinate both channels to achieve robust self-evolution. We present COVE, a unified agent self-evolution framework that combines harness-based and parameter-based learning through task-aware routing, stage-aware scheduling, and knowledge optimization. Through this design, COVE treats self-evolution not as indiscriminate accumulation of experience, but as a coordinated process that matches tasks and knowledge types to appropriate learning mechanisms. Experiments across multiple task categories show that COVE outperforms single-channel evolution strategies, demonstrating more robust and efficient improvement under changing environments.

Problem

• self-evolution
• adaptive memory
• parameter coordination
• large language models
• dynamic environments

Innovation

• adaptive memory-parameter coordination
• self-evolution
• task-aware routing
• stage-aware scheduling
• LLM agents