MindForge Server is an advanced AI-powered backend system featuring a sophisticated multi-agent architecture designed for intelligent personal knowledge management, journaling, and conversational AI. Built with Node.js, MongoDB, and integrated with multiple AI services, it provides a comprehensive platform for users to interact with AI agents that can understand, learn, and respond with context-aware intelligence.
- 9 Specialized AI Agents working in coordinated workflows
- Dynamic Agent Orchestration with supervisor-driven task allocation
- Quality Assurance through monitoring and iterative improvement
- Real-time Streaming responses with workflow transparency
- RAG (Retrieval-Augmented Generation) memory system
- 7 Memory Types: User preferences, behavioral patterns, emotional patterns, topics of interest, goals, personal insights, conversation context
- Advanced Search with relevance scoring and pattern recognition
- Persistent Learning across conversations and sessions
- Personal Journaling with versioning and tagging
- Conversational History with context preservation
- User Authentication via Clerk integration
- Real-time Updates with Redis caching
-
Supervisor Agent (
supervisor)- Role: Central coordinator and workflow orchestrator
- Responsibilities: Analyzes requests, plans responses, coordinates other agents
- Output: JSON-structured task assignments and mission coordination
-
Retrieval Agent (
retrieval)- Role: Information gathering and context search
- Capabilities: Searches journals, memories, chat history for relevant context
- Integration: RAG memory system and journal database
-
Memory Agent (
memory)- Role: Long-term context and pattern management
- Functions: Creates, updates, and manages user memories and insights
- Types: Handles 7 different memory categories for comprehensive user modeling
-
Tags Agent (
tags)- Role: Categorization and metadata generation
- Output: Contextual tags in format
#tagname(clean, no quotes/backticks) - Integration: Automatic tag application to journals and memories
-
Emotion Agent (
emotion)- Role: Emotional intelligence and sentiment analysis
- Capabilities: Analyzes emotional context, provides empathetic responses
- Pattern Recognition: Identifies emotional triggers and behavioral patterns
-
Enhancement Agent (
enhancement)- Role: Content improvement and enrichment
- Functions: Improves response quality, suggests alternatives, adds insights
- Optimization: Enhances user experience through iterative improvements
-
Summarization Agent (
summarization)- Role: Information consolidation and structure
- Capabilities: Synthesizes information, creates structured overviews
- Processing: Handles complex information organization
-
Report Agent (
report)- Role: Structured response generation and analysis
- Output: Comprehensive, well-structured reports and insights
- Analytics: Progress tracking and pattern analysis
-
Monitor Agent (
monitor)- Role: Quality assurance and system monitoring
- Functions: Evaluates response quality, ensures consistency
- Scoring: 1-10 satisfaction scoring with improvement feedback
graph TD
A[User Request] --> B[Supervisor Agent]
B --> C{Analyze & Plan}
C --> D[Select Agent]
D --> E[Execute Task]
E --> F[Monitor Quality]
F --> G{Quality Score ≥ 7?}
G -->|No| H[Improve Response]
H --> E
G -->|Yes| I[Continue Workflow]
I --> J{Mission Complete?}
J -->|No| D
J -->|Yes| K[Final Response]
- Node.js with Express.js
- MongoDB with Mongoose ODM
- Redis for caching and session management
- Clerk for authentication and user management
- Moonshot/Kimi API for LLM processing
- OpenAI Agents Framework for multi-agent coordination
- Custom Agent System with specialized prompts and workflows
- Streaming Responses with Server-Sent Events (SSE)
- CRUD Operations: Create, read, update, delete journals
- Versioning System: Track journal history and changes
- Tag Management: Add/remove tags from journal entries
- Search Functionality: Keyword and tag-based search
- Bulk Operations: Multi-journal updates and analysis
- Real-time Conversations with AI agents
- Context Preservation across sessions
- Journal Integration with reference linking
- Streaming Responses with workflow visibility
- History Management with Redis caching
- Memory CRUD: Create, read, update, delete memories
- Advanced Search: Query-based memory retrieval
- Type Management: 7 specialized memory categories
- Statistics: Memory usage analytics and insights
- Tag Integration: Memory categorization and organization
{
firstName: String,
lastName: String,
username: String (required),
email: String (required, unique),
clerkId: String (required, unique),
avatarUrl: String,
journalIds: [ObjectId],
chatIds: [ObjectId],
summary: String,
insight: String
}{
title: String (required),
content: String,
userId: ObjectId (required),
tags: [String],
audioIds: [String],
nonTitleUpdatedAt: Date,
timestamps: true
}{
userId: ObjectId (required),
memoryType: Enum (7 types),
title: String (required),
content: String (required),
metadata: {
sourceType: String,
sourceId: String,
confidence: Number (0-1),
relevanceScore: Number (0-1),
lastAccessedAt: Date,
accessCount: Number
},
tags: [String],
embeddings: {
vector: [Number],
model: String
},
isActive: Boolean,
relatedMemories: [ObjectId]
}The system includes 26 specialized tools for AI agents:
create_journal,get_journal,update_journal,delete_journalsearch_journals,add_tags,remove_tagsget_journal_versions,set_journal_version,rename_journalget_journal_history
create_memory,get_user_memories,get_memory,update_memorysearch_memories,get_memories_by_typeadd_memory_tags,remove_memory_tags,delete_memoryget_memory_stats
create_chat,get_chat,get_chat_historyupdate_chat_name,delete_chat
- Dual Response Mode: JSON for agent-to-agent communication, clean text for backend/user
- Tag Formatting: Automatic cleanup of
#tagswithout backticks or quotes - Tool Execution: Immediate tool execution without permission requests
- Quality Monitoring: 1-10 satisfaction scoring with iterative improvement
- Test Mode Bypass: Internal tool calls with authentication bypass
- Streaming Workflows: Real-time visibility into agent coordination
- Memory Persistence: Long-term user understanding and pattern recognition
- Context Integration: Journal references and selected text emphasis
- Intelligent Journaling with AI-powered insights
- Conversational AI with persistent memory
- Personal Growth Tracking through pattern analysis
- Context-Aware Responses based on historical data
- Modular Architecture with specialized agents
- Extensible Tool System for new capabilities
- Comprehensive API with full CRUD operations
- Real-time Monitoring and quality assurance
- Clerk Integration for secure user authentication
- Test Mode Security for internal tool execution
- User Isolation with strict data access controls
- API Rate Limiting and error handling
- Redis Caching for conversation history
- Streaming Responses for better user experience
- Agent Timeouts and retry mechanisms
- Optimized Database queries with indexing
- Dynamic Workflow Coordination with iterative task management
- Quality-Driven Responses with satisfaction scoring
- Persistent Learning through memory accumulation
- Pattern Recognition across user behaviors and preferences
- Context-Aware Intelligence with historical understanding
- Node.js 18+
- MongoDB 6+
- Redis 6+
- Clerk account
- Moonshot/Kimi API key
# Clone repository
git clone https://github.com/MindForge-AdventureX2025/MindForgeServer.git
# Install dependencies
cd MindForgeServer
npm install
# Configure environment variables
cp .env.example .env
# Edit .env file with your configuration
# Start development server
npm run dev
# Production start
npm start# Database
MONGO_URI=mongodb://localhost:27017/mindforge
REDIS_URL=redis://localhost:6379
# AI Services
MOONSHOT_API_KEY=your_moonshot_api_key
MOONSHOT_BASE_URL=https://api.moonshot.cn/v1
# Authentication
CLERK_PUBLISHABLE_KEY=your_clerk_key
CLERK_SECRET_KEY=your_clerk_secret
# Server
PORT=3000
CORS_ORIGIN=*MindForge服务器是一个先进的AI驱动后端系统，具有复杂的多智能体架构，专为智能个人知识管理、日记记录和对话AI而设计。基于Node.js、MongoDB构建，并集成多个AI服务，提供了一个全面的平台，让用户与能够理解、学习并以上下文感知智能响应的AI智能体进行交互。
- 9个专业AI智能体协调工作流程
- 动态智能体编排，由监督者驱动任务分配
- 质量保证，通过监控和迭代改进
- 实时流式响应，工作流透明化
- **RAG（检索增强生成）**内存系统
- 7种内存类型：用户偏好、行为模式、情感模式、兴趣话题、目标、个人洞察、对话上下文
- 高级搜索，具有相关性评分和模式识别
- 持久学习，跨对话和会话
- 个人日记，支持版本控制和标签
- 对话历史，上下文保持
- 用户认证，通过Clerk集成
- 实时更新，使用Redis缓存
-
监督智能体 (
supervisor)- 角色：中央协调器和工作流编排者
- 职责：分析请求、规划响应、协调其他智能体
- 输出：JSON结构化任务分配和任务协调
-
检索智能体 (
retrieval)- 角色：信息收集和上下文搜索
- 能力：搜索日记、记忆、聊天历史以获取相关上下文
- 集成：RAG记忆系统和日记数据库
-
记忆智能体 (
memory)- 角色：长期上下文和模式管理
- 功能：创建、更新和管理用户记忆和洞察
- 类型：处理7种不同的记忆类别，进行全面的用户建模
-
标签智能体 (
tags)- 角色：分类和元数据生成
- 输出：格式为
#tagname的上下文标签（干净，无引号/反引号） - 集成：自动将标签应用到日记和记忆
-
情感智能体 (
emotion)- 角色：情感智能和情感分析
- 能力：分析情感上下文，提供共情响应
- 模式识别：识别情感触发器和行为模式
-
增强智能体 (
enhancement)- 角色：内容改进和丰富
- 功能：改进响应质量，建议替代方案，添加洞察
- 优化：通过迭代改进增强用户体验
-
总结智能体 (
summarization)- 角色：信息整合和结构化
- 能力：综合信息，创建结构化概述
- 处理：处理复杂信息组织
-
报告智能体 (
report)- 角色：结构化响应生成和分析
- 输出：全面、结构良好的报告和洞察
- 分析：进度跟踪和模式分析
-
监控智能体 (
monitor)- 角色：质量保证和系统监控
- 功能：评估响应质量，确保一致性
- 评分：1-10满意度评分，提供改进反馈
graph TD
A[用户请求] --> B[监督智能体]
B --> C{分析和规划}
C --> D[选择智能体]
D --> E[执行任务]
E --> F[监控质量]
F --> G{质量评分 ≥ 7?}
G -->|否| H[改进响应]
H --> E
G -->|是| I[继续工作流]
I --> J{任务完成?}
J -->|否| D
J -->|是| K[最终响应]
- Node.js 配合 Express.js
- MongoDB 使用 Mongoose ODM
- Redis 用于缓存和会话管理
- Clerk 用于认证和用户管理
- Moonshot/Kimi API 用于LLM处理
- OpenAI智能体框架 用于多智能体协调
- 自定义智能体系统 具有专业提示和工作流
- 流式响应 使用服务器发送事件（SSE）
- CRUD操作：创建、读取、更新、删除日记
- 版本系统：跟踪日记历史和变更
- 标签管理：添加/删除日记条目标签
- 搜索功能：基于关键词和标签的搜索
- 批量操作：多日记更新和分析
- 实时对话 与AI智能体
- 上下文保持 跨会话
- 日记集成 支持引用链接
- 流式响应 工作流可见性
- 历史管理 使用Redis缓存
- 记忆CRUD：创建、读取、更新、删除记忆
- 高级搜索：基于查询的记忆检索
- 类型管理：7个专业记忆类别
- 统计信息：记忆使用分析和洞察
- 标签集成：记忆分类和组织
系统包括26个专业工具供AI智能体使用：
create_journal,get_journal,update_journal,delete_journalsearch_journals,add_tags,remove_tagsget_journal_versions,set_journal_version,rename_journalget_journal_history
create_memory,get_user_memories,get_memory,update_memorysearch_memories,get_memories_by_typeadd_memory_tags,remove_memory_tags,delete_memoryget_memory_stats
create_chat,get_chat,get_chat_historyupdate_chat_name,delete_chat
- 双响应模式：智能体间通信使用JSON，后端/用户使用干净文本
- 标签格式化：自动清理
#tags，无反引号或引号 - 工具执行：立即执行工具，无需请求权限
- 质量监控：1-10满意度评分，迭代改进
- 测试模式绕过：内部工具调用，绕过认证
- 流式工作流：智能体协调的实时可见性
- 记忆持久化：长期用户理解和模式识别
- 上下文集成：日记引用和选定文本强调
- 智能日记 AI驱动的洞察
- 对话AI 持久记忆
- 个人成长跟踪 通过模式分析
- 上下文感知响应 基于历史数据
- 模块化架构 专业智能体
- 可扩展工具系统 新功能支持
- 全面API 完整CRUD操作
- 实时监控 质量保证
- Clerk集成 安全用户认证
- 测试模式安全 内部工具执行
- 用户隔离 严格数据访问控制
- API速率限制 错误处理
- Redis缓存 对话历史
- 流式响应 更好的用户体验
- 智能体超时 重试机制
- 优化数据库 索引查询
- 动态工作流协调 迭代任务管理
- 质量驱动响应 满意度评分
- 持久学习 记忆积累
- 模式识别 跨用户行为和偏好
- 上下文感知智能 历史理解
- Node.js 18+
- MongoDB 6+
- Redis 6+
- Clerk账户
- Moonshot/Kimi API密钥
# 克隆仓库
git clone https://github.com/MindForge-AdventureX2025/MindForgeServer.git
# 安装依赖
cd MindForgeServer
npm install
# 配置环境变量
cp .env.example .env
# 编辑.env文件，填入您的配置
# 启动开发服务器
npm run dev
# 生产环境启动
npm start# 数据库
MONGO_URI=mongodb://localhost:27017/mindforge
REDIS_URL=redis://localhost:6379
# AI服务
MOONSHOT_API_KEY=your_moonshot_api_key
MOONSHOT_BASE_URL=https://api.moonshot.cn/v1
# 认证
CLERK_PUBLISHABLE_KEY=your_clerk_key
CLERK_SECRET_KEY=your_clerk_secret
# 服务器
PORT=3000
CORS_ORIGIN=*完整的API文档和使用示例请参考项目中的测试文件和控制器实现。
欢迎贡献代码！请查看项目的贡献指南和代码规范。
本项目采用Apache-2.0许可证。详情请查看LICENSE文件。
- 作者: HaolongChen
- 许可证: Apache-2.0
- 仓库: https://github.com/MindForge-AdventureX2025/MindForgeServer
- 版本: 1.0.0
后端框架: Node.js + Express.js
数据库: MongoDB + Mongoose
缓存: Redis
认证: Clerk
AI服务: Moonshot/Kimi API, OpenAI Agents
实时通信: Server-Sent Events (SSE)
安全: Helmet, CORS
MindForge Server - 下一代AI驱动的个人知识管理平台 🚀