---
id: 20260612-T0-15
title: "CWL：让长周期AI代理拥有无限上下文窗口"
title_en: "CWL gives long-horizon agents unlimited context"
url: https://ai.daily.yangsir.net/daily/20260612-T0-15
issue_date: 2026-06-12
publish_date: 2026-06-11T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2606.11213
---

# CWL：让长周期AI代理拥有无限上下文窗口

卡内基梅隆大学发布上下文生命周期管理（CWL），解决长周期AI代理的上下文瓶颈。CWL通过分层语义保留和智能淘汰机制，实现预算内无限历史积累。在机器人任务中，性能提升35%。支持多轮复杂推理，适用于长期对话系统。

## English Version

**CWL gives long-horizon agents unlimited context**

Carnegie Mellon released Context Window Lifecycle (CWL) to solve context bottlenecks for long-horizon AI agents. CWL uses layered semantic retention and intelligent eviction for unlimited in-budget history. Shows 35% performance improvement in robotics tasks. Supports multi-round complex reasoning for long-term dialogue systems.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2606.11213)

**详情页**：https://ai.daily.yangsir.net/daily/20260612-T0-15

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*