DocMind Architecture Visualization
Python 3.11+ TypeScript 5.x Docker + K8s
DocMind
企业级 RAG 智能知识库系统Enterprise RAG Knowledge Base System

全栈 AI 知识库系统，支持 RAG 检索增强生成、文档管理、混合搜索、流式问答、自主 Agent 工具调用。 Full-stack AI knowledge base system with RAG (Retrieval-Augmented Generation), document management, hybrid search, streaming Q&A, and an autonomous Agent with tool calling.

25K+
代码行数Lines of Code
15
API 模块API Modules
12
数据模型Data Models
11
Agent 工具Agent Tools
311
自动化测试Total Tests
12
优化轮次Optimization Rounds
FastAPI SQLAlchemy 2.0 Vue 3 TypeScript Naive UI MySQL 8 Redis Elasticsearch 8 Kafka MinIO DeepSeek API LangChain Pinia Vue Router Prometheus Grafana
A

系统架构System Architecture

多层架构，关注点分离：表现层（Vue 3）→ API 层（FastAPI）→ 业务层（Services）→ 数据层（Models + Infrastructure）。 Multi-layered architecture with clear separation of concerns: Presentation Layer (Vue 3) -> API Layer (FastAPI) -> Business Layer (Services) -> Data Layer (Models + Infrastructure).

CLIENT LAYER / 客户端层
Vue 3 SPA
Naive UI Components
Pinia State Management
Vue Router
SSE / WebSocket Client
Axios HTTP Client
HTTP / WebSocket / SSE
API LAYER / API 网关层
Auth Router
Chat Router
Documents Router
Knowledge Router
Agent Router
Monitoring Router
Workflow Router
+ 7 more routers
Dependency Injection / Service Layer
BUSINESS LAYER / 业务逻辑层
RAG Pipeline
Agent Engine
Document Parser
Embedding Service
Knowledge Service
Workflow Engine
Graph RAG Service
Memory Service
Semantic Chunker
ORM / SDK / Client Libraries
DATA LAYER / 数据存储层
MySQL 8
Redis
Elasticsearch 8
Kafka
MinIO
Prometheus
Container Network / Docker Compose
INFRASTRUCTURE / 基础设施层
Circuit Breaker
Rate Limiter
Distributed Tracing
Structured Logging
Health Checks
Cache Layer
Kafka Worker
P

后端模块Backend Modules

Python（FastAPI）后端遵循 Router → Service → Model 模式，15 个 API 端点、17 个业务服务、12 个数据模型。 Python (FastAPI) backend following the Router -> Service -> Model pattern. 15 API endpoints, 17 business services, 12 SQLAlchemy models.

E

API Endpoints

backend/app/api/v1/endpoints/
14 FastAPI routers handling REST API, WebSocket, and SSE streaming endpoints.
auth.py chat.py documents.py knowledge.py agent.py monitoring.py workflow.py users.py organizations.py prompts.py manuals.py memory.py files.py notifications.py
S

Business Services

backend/app/services/
Core business logic: authentication, document processing, embeddings, RAG, workflows, and graph-based retrieval.
auth_service.py embedding_service.py document_parser.py knowledge_service.py rag_service.py graph_rag_service.py workflow_engine.py memory_service.py semantic_chunker.py file_service.py masking_service.py permission_service.py audit_service.py organization_service.py
R

RAG Pipeline

backend/app/rag/
Retrieval-Augmented Generation pipeline with hybrid search, reranking, context compression, and evaluation.
pipeline.py retriever.py reranker.py context_window.py context_compressor.py query_processor.py evaluator.py cache.py metrics.py
A

Agent System

backend/app/agent/
ReAct-style autonomous agent with self-registering tools, skill learning, and subagent delegation.
registry.py loop.py context.py skills.py subagent.py tools.py service.py
M

Data Models

backend/app/models/
SQLAlchemy 2.0 models using modern Mapped + mapped_column syntax. 12 domain models with relationships.
user.py chat.py document.py knowledge_job.py organization.py rbac.py prompt.py manual.py workflow.py notification.py user_audit.py
C

Core Infrastructure

backend/app/core/
Infrastructure layer: database, cache, search, messaging, security, observability, and configuration.
database.py redis.py elasticsearch.py kafka_client.py minio_client.py security.py circuit_breaker.py prometheus.py tracing.py logging.py middleware.py config/
V

前端模块Frontend Modules

Vue 3 + TypeScript + Naive UI 前端，20+ 页面组件，Pinia 状态管理，组合式函数，实时流式传输支持。 Vue 3 + TypeScript + Naive UI frontend with 20+ views, Pinia state management, composables, and real-time streaming support.

V

Page Views

frontend/src/views/
20+ page components covering chat, knowledge management, agent, dashboard, monitoring, workflow, and admin features.
chat/ knowledge/ agent/ dashboard/ monitoring/ workflow/ documents/ search/ conversations/ upload/ login/ users/ organizations/ prompts/ manual/ notifications/ profile/ admin/ firsthome/ system-about/
P

State Management

frontend/src/stores/
Pinia stores for global state management: application state, chat sessions, user auth, workflow, and notifications.
app.ts chat.ts user.ts workflow.ts notification.ts
C

Composables

frontend/src/composables/
Reusable composition functions for chat connections, message handling, error management, and data prefetching.
useDebounce.ts useErrorHandler.ts usePrefetch.ts
A

API Client

frontend/src/api/
TypeScript API client modules with typed request/response interfaces for all backend endpoints.
auth.ts chat.ts conversation.ts knowledge.ts agent.ts rag.ts search.ts workflow.ts monitoring.ts user.ts organization.ts prompt.ts manual.ts memory.ts notification.ts audit.ts
R

RAG 检索增强生成流水线RAG Pipeline

智能查询处理：复杂度分类、自适应策略选择、混合搜索（BM25 + 向量）、RRF 融合、重排序、上下文压缩。 Intelligent query processing with complexity classification, adaptive strategy selection, hybrid search (BM25 + Vector), RRF fusion, reranking, and context compression.

1 Query Processing Flow / 查询处理流程
User Query
用户查询
Complexity Classifier
复杂度分类器
Strategy Selection
策略选择
Query Expansion
查询扩展
Strategy: keyword_only | hybrid | hybrid_hyde
2 Hybrid Search / 混合检索
BM25 Keyword Search
关键词搜索
Vector Similarity Search
向量相似度搜索
RRF Fusion
倒数排名融合
Reranker
重排序
Context Window
上下文窗口
3 Generation / 生成阶段
Context Compression
上下文压缩
Citation Tracking
引用追踪
LLM Generation
大模型生成
SSE Stream Response
流式响应
1

Query Complexity Classification / 查询复杂度分类

Classifies queries as simple, moderate, or complex to select optimal retrieval strategy.

query_processor.py
|
2

Adaptive Strategy Selection / 自适应策略选择

keyword_only for simple factual, hybrid for moderate, hybrid_hyde for complex reasoning queries.

pipeline.py
|
3

Hybrid Search + RRF Fusion / 混合检索 + RRF 融合

BM25 keyword search + dense vector search, merged via Reciprocal Rank Fusion for optimal recall.

retriever.py
|
4

Cross-Encoder Reranking / 交叉编码器重排序

Neural reranker scores query-document pairs for precise relevance ranking.

reranker.py
|
5

Context Window + Compression / 上下文窗口 + 压缩

Smart context assembly with token budget management and redundancy removal.

context_window.py / context_compressor.py
|
6

Citation Tracking + LLM Generation / 引用追踪 + 大模型生成

Sources tracked and cited in responses, streamed via SSE for real-time user experience.

evaluator.py / metrics.py
G

Agent 智能体系统Agent System

ReAct 风格自主智能体，自注册工具注册表、并行执行、上下文压缩、技能学习、子智能体委托。 ReAct-style autonomous agent with self-registering tool registry, parallel execution, context compression, skill learning, and subagent delegation.

I Input / 输入

User Query
Natural language question
Context
Conversation history
Metadata
User, org, permissions
|

O Output / 输出

Final Answer
SSE streaming response
Tool Results
Intermediate outputs
Reasoning Trace
Step-by-step log

ReAct Agent Loop

Thought -> Action -> Observation -> Repeat

思考 -> 行动 -> 观察 -> 循环

L Loop Engine / 循环引擎

Max iterations control
Early stopping detection
Error recovery & retry
Token budget management

C Context Engine / 上下文引擎

Context compression
Memory management
Conversation window
Token counting

S Skill Learning / 技能学习

Pattern recognition
Success/failure tracking
Skill composition
Auto-optimization

D SubAgent / 子智能体

Task delegation
Specialized agents
Result aggregation
Hierarchical planning

T Tool Registry (11 tools) / 工具注册表

Document Search
Knowledge Query
Web Search
Code Analysis
Data Analysis
Summarize
Translate
Conversation
Memory Store
Prompt Template
File Operations

R Tool Execution / 工具执行

Parallel execution support
Timeout & retry policies
Result validation
SSE real-time visualization
D

数据流转Data Flow

端到端数据流转：从文档上传到处理流水线再到智能查询响应。 End-to-end data flow from document upload through processing pipeline to intelligent query response.

1 Document Ingestion Flow / 文档摄入流程
File Upload
文件上传
MinIO Storage
对象存储
DB Record
数据库记录
Kafka Message
消息队列
Worker Process
异步处理
Parse & Chunk
解析分块
Embedding
向量化
Elasticsearch
索引存储
2 Query Response Flow / 查询响应流程
User Query
用户查询
WebSocket/SSE
实时连接
RAG Pipeline
检索增强
Hybrid Search
混合搜索
Rerank
重排序
Context Assembly
上下文组装
DeepSeek LLM
大模型推理
Stream Response
流式响应
I

基础设施Infrastructure

生产级基础设施：缓存、消息队列、搜索引擎、对象存储、可观测性。 Production-grade infrastructure with caching, messaging, search, storage, and observability.

D

MySQL 8

Primary relational database. 12 SQLAlchemy 2.0 models with async support, connection pooling, and migration management.

R

Redis

In-memory cache for sessions, rate limiting, token blacklists, and distributed locks. Pub/Sub for real-time notifications.

E

Elasticsearch 8

Hybrid search engine supporting BM25 keyword search and dense vector similarity search with RRF fusion.

K

Kafka

Async message queue for document processing pipeline. Decouples upload from heavy parsing and embedding operations.

M

MinIO

S3-compatible object storage for document files. Supports multiple file formats: PDF, DOCX, TXT, MD, HTML.

P

Prometheus + Grafana

17+ RAG metrics, 18 Grafana dashboard panels. Custom metrics for latency, throughput, cache hits, and quality scores.

S

安全防护层Security Layers

纵深防御安全架构：认证、授权、输入验证、基础设施加固。 Defense-in-depth security architecture with authentication, authorization, input validation, and infrastructure hardening.

1

Authentication / 身份认证

JWT token-based authentication with access/refresh token rotation and token blacklisting on logout.

JWT Token Blacklist bcrypt Refresh Rotation
2

Authorization / 权限控制

Role-Based Access Control (RBAC) with organization-level multi-tenancy and granular permission management.

RBAC Multi-Tenancy Permission Check
3

Input Validation / 输入验证

Pydantic schema validation, SSRF prevention, path traversal protection, and AST-based code sandboxing.

Pydantic SSRF Prevention Path Traversal AST Sandbox
4

Rate Limiting & Protection / 限流防护

Redis-based rate limiting, brute force protection, CORS policy, and request throttling per user/IP.

Rate Limiting Brute Force CORS Throttling
5

Data Protection / 数据保护

Sensitive data masking, encrypted storage, audit logging, and secure file handling with content type validation.

Data Masking Audit Log Content Validation
T

测试与质量Testing & Quality

全面测试覆盖，311 个测试用例，12 轮优化迭代，生产级监控。 Comprehensive test coverage with 311 total tests, 12 optimization rounds, and production-grade monitoring.

216
后端测试Backend Tests
95
前端测试Frontend Tests
12
优化轮次Optimization Rounds
100%
重构完成Refactor Complete
M

17+ RAG 指标17+ RAG Metrics

Prometheus 指标覆盖延迟、吞吐量、缓存命中率、检索质量、重排序分数和生成指标。Prometheus metrics covering latency, throughput, cache hit rate, retrieval quality, reranker scores, and generation metrics.

G

18 个 Grafana 面板18 Grafana Panels

全面的仪表盘：系统健康、RAG 管线性能、API 延迟分布和错误率追踪。Comprehensive dashboards for system health, RAG pipeline performance, API latency distribution, and error rate tracking.

R

52/53 重构任务52/53 Refactor Tasks

系统化代码质量改进：服务提取、异常处理、类型安全、性能优化和文档完善。Systematic code quality improvements: service extraction, error handling, type safety, performance optimization, and documentation.