https://docs.nbagent.ai/static/js/async/148.4ec3a13a4f.js

1"use strict";(self.rspackChunknebulaAiDocs=self.rspackChunknebulaAiDocs||[]).push([["148"],{4319(n,e,r){r.r(e),r.d(e,{default:()=>d});var s=r(2615),i=r(416);function l(n){let e={a:"a",em:"em",h1:"h1",h2:"h2",h3:"h3",hr:"hr",img:"img",li:"li",p:"p",strong:"strong",table:"table",tbody:"tbody",td:"td",th:"th",thead:"thead",tr:"tr",ul:"ul",...(0,i.R)(),...n.components};return(0,s.jsxs)(s.Fragment,{children:[(0,s.jsxs)(e.h1,{id:"3系统架构",children:[(0,s.jsx)(e.a,{className:"rp-header-anchor","aria-hidden":"true",href:"#3系统架构",children:"#"}),"3、系统架构"]}),"\n",(0,s.jsx)(e.p,{children:(0,s.jsx)(e.img,{src:"/images/architecture/1765800832015-c3a74ef5-0a1c-4188-ac96-a44e5ef3a102.png",alt:"NebulaAI 系统架构图"})}),"\n",(0,s.jsxs)(e.h2,{id:"1-架构总览-system-architecture-overview",children:[(0,s.jsx)(e.a,{className:"rp-header-anchor","aria-hidden":"true",href:"#1-架构总览-system-architecture-overview",children:"#"}),"1. 架构总览 (System Architecture Overview)"]}),"\n",(0,s.jsxs)(e.p,{children:["本系统（NebulaAI）采用 ",(0,s.jsx)(e.strong,{children:"“端-云-脑”一体化"})," 的分层微服务架构。系统自上而下分为 ",(0,s.jsx)(e.strong,{children:"应用层（交互与生态）"}),"、",(0,s.jsx)(e.strong,{children:"服务层（支撑与中间件）"}),"、",(0,s.jsx)(e.strong,{children:"智能体底座（核心认知引擎）"})," 以及 ",(0,s.jsx)(e.strong,{children:"LLM 接口层（模型算力）"}),"。该架构通过高内聚低耦合的设计，实现了从用户交互前端（React/Next.js）到异构大模型后端（Python/Hybrid LLMs）的全链路打通，确保了平台的高扩展性与企业级数据一致性。"]}),"\n",(0,s.jsx)(e.hr,{}),"\n",(0,s.jsxs)(e.h2,{id:"2-详细分层设计",children:[(0,s.jsx)(e.a,{className:"rp-header-anchor","aria-hidden":"true",href:"#2-详细分层设计",children:"#"}),"2. 详细分层设计"]}),"\n",(0,s.jsxs)(e.h3,{id:"21-应用层-application-layer--交互中心与生态构建",children:[(0,s.jsx)(e.a,{className:"rp-header-anchor","aria-hidden":"true",href:"#21-应用层-application-layer--交互中心与生态构建",children:"#"}),"2.1 应用层 (Application Layer) —— 交互中心与生态构建"]}),"\n",(0,s.jsx)(e.p,{children:(0,s.jsxs)(e.em,{children:["本层基于 ",(0,s.jsx)(e.strong,{children:"Next.js/React"})," 构建，负责具体业务场景的呈现与人机交互，通过组件化设计支持灵活的功能扩展。"]})}),"\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"生态与管理 (App Store & Management)"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"应用商店 & 插件中心"}),": 建立基于 React 组件的开放市场，允许开发者通过标准化接口发布 Agent 技能与前端 UI 组件。"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"应用管理 & 用户管理"}),": 提供多租户管理后台，基于 ",(0,s.jsx)(e.strong,{children:"MySQL"})," 存储企业组织架构、角色权限（RBAC）及计费策略。"]}),"\n"]}),"\n"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"交互形式 (Smart Interface)"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"智能卡片 (Smart Cards)"}),": 利用 Next.js 的服务端渲染（SSR）能力，动态生成可视化数据卡片（如报表、图表），超越纯文本对话体验。"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"工作流 (Workflows)"}),": 提供低代码（Low-Code）可视化画布，允许用户通过拖拽编排 Agent 的任务链路。"]}),"\n"]}),"\n"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"业务逻辑扩展"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"云函数 (Cloud Functions)"}),": 集成 Python Serverless 运行时，支持用户编写自定义 Python 脚本作为 Agent 的动态能力扩展。"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"长期记忆 (Long-term Memory UI)"}),": 可视化展示用户的历史画像与偏好设置。"]}),"\n"]}),"\n"]}),"\n"]}),"\n",(0,s.jsxs)(e.h3,{id:"22-服务层-service-layer--基础设施与中间件",children:[(0,s.jsx)(e.a,{className:"rp-header-anchor","aria-hidden":"true",href:"#22-服务层-service-layer--基础设施与中间件",children:"#"}),"2.2 服务层 (Service Layer) —— 基础设施与中间件"]}),"\n",(0,s.jsx)(e.p,{children:(0,s.jsxs)(e.em,{children:["本层作为系统的稳定支撑层，通过 ",(0,s.jsx)(e.strong,{children:"Python 后端服务"})," 提供通用能力，确保系统的高可用与安全性。"]}

1)}),"\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"安全与通信"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"鉴权 (Auth)"}),": 统一网关层，支持 JWT/OAuth2.0 认证，确保 API 调用与 WebSocket 连接的安全性。"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"API 调用 & Mock 服务"}),": 标准化 RESTful/gRPC 接口封装，支持第三方系统集成，并提供 Mock 能力以便于前后端并行开发与测试。"]}),"\n"]}),"\n"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"多模态支持"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"音视频 (Audio/Video)"}),": 处理语音转文字（STT）与文字转语音（TTS）流，支持多模态输入输出。"]}),"\n"]}),"\n"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"高性能数据基座"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"缓存与存储 (Cache & Storage)"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:["采用 ",(0,s.jsx)(e.strong,{children:"Redis"})," 缓存会话热数据（Session State）与 API 响应，实现毫秒级响应。"]}),"\n",(0,s.jsxs)(e.li,{children:["采用 ",(0,s.jsx)(e.strong,{children:"MySQL"})," 实现业务数据的持久化存储，保障数据强一致性（ACID）。"]}),"\n"]}),"\n"]}),"\n"]}),"\n"]}),"\n"]}),"\n",(0,s.jsxs)(e.h3,{id:"23-智能体底座-agent-foundation--核心认知引擎",children:[(0,s.jsx)(e.a,{className:"rp-header-anchor","aria-hidden":"true",href:"#23-智能体底座-agent-foundation--核心认知引擎",children:"#"}),"2.3 智能体底座 (Agent Foundation) —— 核心认知引擎"]}),"\n",(0,s.jsx)(e.p,{children:(0,s.jsxs)(e.em,{children:["本层是 NebulaAI 的“大脑”，由 ",(0,s.jsx)(e.strong,{children:"Python"})," 深度开发，集成了思维链（CoT）与向量检索技术。"]})}),"\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"认知调度"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"智能编排 (Orchestration)"}),": 基于 ReAct 或 TOT（Tree of Thoughts）范式，分析用户意图，动态拆解任务步骤。"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"执行计划 (Planning)"}),": 自动生成分步执行路径，并在每一步执行后进行结果验证与自我修正。"]}),"\n"]}),"\n"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"记忆与知识增强 (Memory & RAG)"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"RAG 检索"}),": 集成 ",(0,s.jsx)(e.strong,{children:"FAISS"})," 向量数据库，构建高维稠密检索系统。对企业文档进行切片 Embedding，实现私有领域知识的精准召回。"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"核心记忆 (Core Memory)"}),": 维护短期上下文窗口，动态压缩与剪枝（Pruning）对话历史，确保 Prompt 不超限且保留关键信息。"]}),"\n"]}),"\n"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"能力扩展"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"工具注册 (Tool Registration)"}),": 定义标准化的 Python Decorators，自动将本地函数、API 接口转化为 LLM 可理解和调用的工具（Tools）。"]}),"\n"]}),"\n"]}),"\n"]}),"\n",(0,s.jsxs)(e.h3,{id:"24-模型接口层-openai--llm-interface--异构模型网关",children:[(0,s.jsx)(e.a,{className:"rp-header-anchor","aria-hidden":"true",href:"#24-模型接口层-openai--llm-interface--异构模型网关",children:"#"}),"2.4 模型接口层 (OpenAI / LLM Interface) —— 异构模型网关"]}),"\n",(0,s.jsx)(e.p,{children:(0,s.jsxs)(e.em,{children:["本层实现了“模型中立”策略，通过 ",(0,s.jsx)(e.strong,{children:"Model Gateway"})," 统一封装，屏蔽了底层各厂商的技术差异。"]})}),"\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"统一协议 (Unified Interface)"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:["基于 ",(0,s.jsx)(e.strong,{children:"OpenAI SDK 标准格式"})," 封装所有模型接口，支持流式输出（Streaming），便于上层业务无缝切换模型。"]}),"\n"]}),"\n"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"混合模型调度 (Hybrid Routing)"}),":\n",(0,s.jsxs)(e.ul,{children:["\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"本地私有化 (Local/Edge)"}),": 对接 ",(0,s.jsx)(e.strong,{children:"Ollama"})," 运行 Llama3/Mistral 等模型，用于处理隐私敏感数据或高频低成本任务。"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"云端高性能 (Cloud SaaS)"}),": 路由复杂逻辑任务至 ",(0,s.jsx)(e.strong,{children:"Azure OpenAI"}),"、",(0,s.jsx)(e.strong,{children:"OpenAI (GPT-4)"})," 或 ",(0,s.jsx)(e.strong,{children:"Claude"}),"。"]}),"\n",(0,s.jsxs)(e.li,{children:[(0,s.jsx)(e.strong,{children:"国äº

1§åŒ–合规 (China Native)"}),": 深度适配 ",(0,s.jsx)(e.strong,{children:"Qwen (通义千问)"}),"、",(0,s.jsx)(e.strong,{children:"Qianfan (百度千帆)"}),"、",(0,s.jsx)(e.strong,{children:"讯飞星火"}),"，满足国内业务合规与中文理解需求。"]}),"\n"]}),"\n"]}),"\n"]}),"\n",(0,s.jsx)(e.hr,{}),"\n",(0,s.jsxs)(e.h2,{id:"3-技术栈映射摘要",children:[(0,s.jsx)(e.a,{className:"rp-header-anchor","aria-hidden":"true",href:"#3-技术栈映射摘要",children:"#"}),"3. 技术栈映射摘要"]}),"\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",(0,s.jsxs)(e.table,{children:[(0,s.jsx)(e.thead,{children:(0,s.jsxs)(e.tr,{children:[(0,s.jsx)(e.th,{children:(0,s.jsx)(e.strong,{children:"架构层级"})}),(0,s.jsx)(e.th,{children:(0,s.jsx)(e.strong,{children:"图示模块映射"})}),(0,s.jsx)(e.th,{children:(0,s.jsx)(e.strong,{children:"核心技术栈"})})]})}),(0,s.jsxs)(e.tbody,{children:[(0,s.jsxs)(e.tr,{children:[(0,s.jsx)(e.td,{children:(0,s.jsx)(e.strong,{children:"应用层"})}),(0,s.jsx)(e.td,{children:"插件管理、应用商店、智能卡片、工作流"}),(0,s.jsxs)(e.td,{children:[(0,s.jsx)(e.strong,{children:"Next.js / React"})," (SSR, Streaming UI)"]})]}),(0,s.jsxs)(e.tr,{children:[(0,s.jsx)(e.td,{children:(0,s.jsx)(e.strong,{children:"服务层"})}),(0,s.jsx)(e.td,{children:"鉴权、API 调用、缓存与存储"}),(0,s.jsxs)(e.td,{children:[(0,s.jsx)(e.strong,{children:"Python"})," (Backend Framework), ",(0,s.jsx)(e.strong,{children:"Redis"})," (Cache), ",(0,s.jsx)(e.strong,{children:"MySQL"})," (DB)"]})]}),(0,s.jsxs)(e.tr,{children:[(0,s.jsx)(e.td,{children:(0,s.jsx)(e.strong,{children:"底座层"})}),(0,s.jsx)(e.td,{children:"RAG 检索、核心记忆、智能编排"}),(0,s.jsxs)(e.td,{children:[(0,s.jsx)(e.strong,{children:"Python"})," (Logic), ",(0,s.jsx)(e.strong,{children:"FAISS"})," (Vector DB), LangChain/Self-developed Core"]})]}),(0,s.jsxs)(e.tr,{children:[(0,s.jsx)(e.td,{children:(0,s.jsx)(e.strong,{children:"模型层"})}),(0,s.jsx)(e.td,{children:"LLM Interface、Qwen/Ollama/Azure"}),(0,s.jsxs)(e.td,{children:[(0,s.jsx)(e.strong,{children:"Model Gateway"})," (Python Adapters for ",(0,s.jsx)(e.strong,{children:"OpenAI/Azure/Ollama/Qwen/etc."}),")"]})]})]})]}),"\n",(0,s.jsxs)(e.p,{children:["这个架构描述完美契合了 ",(0,s.jsx)(e.strong,{children:"NebulaAI"})," 的图片结构，展现了从底层数据处理、模型推理到上层应用分发的完整技术图谱，既体现了系统的复杂度和专业性，又突出了您选用的技术栈的先进性。"]})]})}function d(){let n=arguments.length>0&&void 0!==arguments[0]?arguments[0]:{},{wrapper:e}={...(0,i.R)(),...n.components};return e?(0,s.jsx)(e,{...n,children:(0,s.jsx)(l,{...n})}):l(n)}d.__RSPRESS_PAGE_META={},d.__RSPRESS_PAGE_META["guide%2Fstart%2Fsystem-architecture.md"]={toc:[{id:"1-架构总览-system-architecture-overview",text:"1. 架构总览 (System Architecture Overview)",depth:2},{id:"2-详细分层设计",text:"2. 详细分层设计",depth:2},{id:"21-应用层-application-layer--交互中心与生态构建",text:"2.1 应用层 (Application Layer) —— 交互中心与生态构建",depth:3},{id:"22-服务层-service-layer--基础设施与中间件",text:"2.2 服务层 (Service Layer) —— 基础设施与中间件",depth:3},{id:"23-智能体底座-agent-foundation--核心认知引擎",text:"2.3 智能体底座 (Agent Foundation) —— 核心认知引擎",depth:3},{id:"24-模型接口层-openai--llm-interface--异构模型网关",text:"2.4 模型接口层 (OpenAI / LLM Interface) —— 异构模型网关",depth:3},{id:"3-技术栈映射摘要",text:"3. 技术栈映射摘要",depth:2}],title:"3、系统架构",headingTitle:"3、系统架构",frontmatter:{}}}}]);