Long-horizon large language model (LLM) agents commonly retain their complete interaction history until compaction is triggered at a predefined threshold. We…
机构:Harvard University
来源:arXiv 2609.35540 | AI4Papers 论文推荐平台