Verified65% confidenceFactTime unknown
传统RAG的流程是单向固定的一次性流程，而Agentic RAG是循环迭代的多步执行流程
3
Sources
65%
Confidence
Long-term
Relevance
6/3/2026
First Seen
Sources
从传统RAG到Agentic RAG：核心原理与实现详解
bilibiliAI研究所-
Related Entities
Related Claims
UnverifiedRAG遵循「查询-检索-生成」的单轮线性范式，前后轮次之间缺乏状态传递机制，无法主动规划多步骤任务66% similarUnverified工作流文档配合提示词后追机制本质上是一种检索增强生成（RAG）的简化实现，用于在上下文重置后快速恢复Agent的任务认知60% similarUnverified拉尔夫循环在技术上属于自回归反馈循环（Auto-regressive Feedback Loop）的一种工程化实现59% similarUnverifiedRAG本质上是单轮、被动的信息检索，无法处理需要多步推理、跨工具协作或动态反馈的复杂任务59% similarUnverifiedThe ReAct pattern's core idea is to have the LLM alternately execute 'Reasoning' and 'Acting' steps in a loop to handle complex multi-step tasks.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/40376API
curl https://kongchang.com/api/v1/knowledge/claims/40376MCP
get_claim(id=40376)