📄 中文摘要
当前的语言模型（LLMs）在设计上是无状态的，每次 API 调用都从零开始，无法记住用户的身份或上下文。OpenClaw 通过使用纯文本文件和快速的语义搜索技术解决了这一问题。该系统将上下文与持久记忆分开，允许记忆以 Markdown 文件的形式存储，并通过 SQLite 数据库进行索引，结合了 BM25 和向量搜索。SOUL.md 文件定义了用户的身份和个性，而 MEMORY.md 文件则存储与用户的交互历史，从而实现了更为个性化的 AI 体验。
OpenClaw 记忆：你的 AI 代理如何真正记住你
出处: OpenClaw Memory: How Your AI Agent Actually Remembers You
发布: 2026年3月31日
📄 English Summary
OpenClaw Memory: How Your AI Agent Actually Remembers You
Current language models (LLMs) are stateless by design, starting from scratch with each API call, which means they cannot remember user identities or context. OpenClaw addresses this issue by utilizing plain text files and fast semantic search technology. The system separates context from persistent memory, allowing memories to be stored as Markdown files indexed in an SQLite database, combining BM25 and vector search. The SOUL.md file defines the user's identity and personality, while the MEMORY.md file stores the history of interactions with the user, enabling a more personalized AI experience.