#MemAgent

MemAgent introduces a reinforcement learning-based memory agent that allows large language models to process ultra-long documents efficiently, maintaining high accuracy with linear computational costs.