text-gradient

AutoMem is a text-gradient recursive self-improvement framework for automated memory architecture search in LLM agents, which discovers task-adaptive architectures that outperform human-designed baselines with improved accuracy and efficiency.