[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"aihot-art-92279":3},{"itemId":4,"vertical":5,"category":6,"source":7,"score":8,"title":9,"summary":10,"analysis":11,"url":12,"coverUrl":13,"direction":13,"marketSignal":13,"publishedAt":14},"92279","ai","技巧与观点","MarkTechPost",61,"Google Cloud 常驻记忆体代理：用连续 LLM 整合替代 RAG 与嵌入","Google Cloud推出常驻记忆体代理，用连续LLM整合替代RAG与嵌入，无需向量数据库。","Google Cloud推出Always-On Memory Agent，通过持续运行的LLM进程管理记忆，无需向量数据库或嵌入，以低延迟低成本实现长期上下文保持。\n· 该代理作为轻量级后台进程24\u002F7运行，基于Google ADK和Gemini 3.1 Flash-Lite，不使用向量数据库或嵌入，而是由LLM读取、思考并将结构化记忆写入SQLite。\n· 架构上，编排器将请求路由至三个子代理：IngestAgent提取摘要、实体、主题和重要性评分；ConsolidateAgent每30分钟运行一次，发现记忆间联系并生成综合摘要；QueryAgent基于所有记忆和综合洞察回答问题，并引用记忆ID。\n· 模型选择针对低延迟和低成本，适合持续后台工作。\n影响\u002F看点：该方案挑战了RAG+嵌入的主流记忆管理范式，通过连续LLM整合实现更智能、更经济的记忆机制，为AI代理的长期上下文保持提供了新思路。","https:\u002F\u002Fwww.marktechpost.com\u002F2026\u002F07\u002F18\u002Fgoogle-clouds-always-on-memory-agent-replaces-rag-and-embeddings-with-continuous-llm-consolidation-on-gemini-3-1-flash-lite\u002F",null,"2026-07-18 15:57:51"]