📄 中文摘要
许多产品声称能够为人工智能代理提供持久记忆，然而这些解决方案都未能解决根本问题。问题不在于代理缺乏存储，而在于缺乏合适的架构。上下文窗口的容量是有限的，许多记忆解决方案将其视为缺陷，而非设计约束。构建最强大代理的团队并不试图让代理记住更多信息，而是致力于让其更好地忘记。记忆解决方案的标准做法通常包括为代理提供数据库访问及存储对话记录等，但这种方法未能有效利用上下文窗口的特性。
上下文窗口是新的记忆架构
出处: The Context Window Is the New Memory Architecture
发布: 2026年3月28日
📄 English Summary
The Context Window Is the New Memory Architecture
Many products claim to provide persistent memory for AI agents, yet these solutions fail to address the core issue. The problem is not that agents lack storage, but that they lack the right architecture. Context windows have finite capacity, and many memory solutions treat this as a bug rather than a design constraint. Teams building the most capable agents are not trying to make them remember more; instead, they focus on making them forget better. The standard approach for memory solutions typically involves giving agents access to a database and storing conversation logs, but this method does not effectively leverage the characteristics of context windows.