Le préprint ExpGraph propose une mémoire en graphe auto-évolutif pour agents LLM

In three linesExpGraph, an arXiv preprint from May 29, 2026, proposes a self-evolving graph memory architecture for LLM agents. The system accumulates experience through dynamic nodes and edges, enabling agents to improve performance on complex tasks without retraining.

Summary generated by Claude — human-verified