1 paper
Songxin Lei, Kun Ouyang, Weilin Ruan +4
Long-horizon LLM agents require memory systems that recover useful evidence from large interaction histories without passing excessive context to downstream models. Existing memory…