# Agent Weaver

**Repository Path**: LambdaH/agent-weaver

## Basic Information

- **Project Name**: Agent Weaver
- **Description**: 智能体工作流核心框架
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-03-30
- **Last Updated**: 2026-04-29

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Agent Weaver

一个面向本地/云主机部署的可视化工作流引擎，完成"信息采集 → 筛选判断 → 总结生成 → 投递沉淀"的完整闭环。

---

## 中文

### 项目简介

Agent Weaver 的核心定位是让用户通过画布式 DAG 编辑器，编排包含 RSS、HTTP、LLM、MCP、Email 在内的多种节点，实现定时运行的自动化信息处理工作流。工作流运行的结果可沉淀到轻量知识库，并通过 **Ask Me** 入口进行基于 RAG 的向量检索、多轮对话与数据可视化回顾。

**技术亮点**：
- **MCP 域化架构**：SSE 客户端兼容 MCP 2024-11-05 协议，LLM 自然语言输出可透明转化为结构化工具调用
- **Ask Me 智能问答**：Embedding 相似度排序 + 可插拔 Skill 路由（股票/天气），一句话触发图表生成与引用溯源
- **工作流执行引擎**：13 种节点类型，支持模板变量 `{{variable}}` 传递、条件分支、定时调度与安全上下文注入
- **工程安全**：JWT 无状态鉴权、AES-GCM 凭证加密、用户数据隔离

### 功能一览

- 登录鉴权与受保护路由（JWT + Spring Security）
- 工作流 CRUD、复制与画布式 DAG 编辑（Vue Flow）
- 13 种节点：Start、HTTP、RSS Feed、Time Filter、Filter List、Condition、Template、Knowledge Write、LLM、LLM Judge、Email、MCP Stock、End
- 节点复制、连线、保存、执行与运行状态高亮
- 定时任务 CRUD（Cron 表达式 + 时区支持）
- 运行历史与节点级输出预览
- 凭证管理（AES-GCM 加密存储）
- 真实 RSS 抓取、DeepSeek API 调用、SMTP 邮件发送
- MCP Stock 节点：新闻冲击模拟、买入/卖出交易、持仓快照
- 工作流知识库沉淀与 Ask Me 对话入口
- 中英双语界面

### 技术栈

| 层级 | 技术 |
|------|------|
| 前端 | Vue 3 + Vite + Vue Router + Vue Flow |
| 后端 | Spring Boot 3 + Java 17 + MyBatis-Plus |
| 数据库 | MySQL 8 |
| 向量检索 | Embedding API（BAAI/bge-large-zh-v1.5）+ 余弦相似度 |
| LLM | DeepSeek API |
| MCP | SSE 客户端 + Python Flask MCP Server |
| 部署 | Docker Compose + Nginx |

### 目录结构

```text
Agent_Weaver/
  Backend/          Spring Boot 后端
  Frontend/         Vue 3 前端
  Database/         数据库初始化、种子数据、快照支持
  mcp-server/       MCP Stock 服务（Flask + SSE）
  deploy/           生产环境部署配置
  docker-compose.local.yml
  docker-compose.prod.yml
```

### 快速开始

#### 本地 Docker 启动

```bash
docker-compose -f docker-compose.local.yml up -d --build
```

本地服务地址：

- 前端：`http://localhost:5173`
- 后端：`http://localhost:8081`
- phpMyAdmin：`http://localhost:8080`
- MySQL：`localhost:3306`
- MCP Stock：`http://localhost:3001`

#### 云主机 / 生产环境

```bash
cp .env.prod.example .env.prod
docker-compose -f docker-compose.prod.yml up -d --build
```

生产环境通过 Nginx Gateway 统一代理 `/api/` 到后端、`/` 到前端。更详细的 Linux 云主机部署步骤请参考 [云主机部署说明.md](./云主机部署说明.md)。

### 数据库初始化

MySQL 启动时按以下优先级初始化：

1. 若 `Database/local-snapshot/` 存在 SQL 快照，优先恢复快照；
2. 否则依次执行 `Database/init/00-create-database.sql` → `01-schema.sql` → `seed/*.sql`。

### 注意事项

- `.env`、`.env.prod` 和本地数据库快照默认被 Git 忽略。
- 生产环境请替换所有默认密码、JWT secret 和加密密钥。
- 首次验证部署链路时，可先用模板值启动，再逐步替换为生产级凭据。

### 许可

本项目采用 [MIT License](./LICENSE)。

---

## English

### Overview

Agent Weaver is a visual workflow engine for local or cloud-VM deployment. It enables users to orchestrate DAG-based workflows with RSS, HTTP, LLM, MCP, and Email nodes, automating the full loop of information collection, filtering, summarization, and delivery. Workflow outputs are persisted to a lightweight knowledge base and can be reviewed through the **Ask Me** entry point, supporting RAG vector retrieval, multi-turn conversations, and data-visualization-driven Q&A.

**Highlights**:
- **MCP Domain Architecture**: A custom SSE client compatible with MCP 2024-11-05 protocol transparently transforms LLM natural-language output into structured tool calls.
- **Ask Me Intelligent Q&A**: Embedding similarity ranking plus pluggable Skill routing (stock / weather) triggers chart generation and citation-backed answers from a single sentence.
- **Workflow Execution Engine**: 13 node types with `{{variable}}` template passing, conditional branching, scheduled execution, and secure-context injection.
- **Engineering Security**: JWT stateless authentication, AES-GCM credential encryption, and user data isolation.

### Features

- Authentication with protected routes (JWT + Spring Security)
- Workflow CRUD, duplication, and canvas-based DAG editing (Vue Flow)
- 13 node types: Start, HTTP, RSS Feed, Time Filter, Filter List, Condition, Template, Knowledge Write, LLM, LLM Judge, Email, MCP Stock, End
- Node copy, connect, save, execute, and run-status highlighting
- Scheduled task CRUD (Cron expressions with timezone support)
- Execution history and per-node output preview
- Credential management (AES-GCM encrypted storage)
- Real RSS fetching, DeepSeek API calls, SMTP email delivery
- MCP Stock node: news-shock simulation, buy/sell trades, portfolio snapshot
- Knowledge-base persistence and Ask Me conversational entry point
- Bilingual UI (Chinese / English)

### Tech Stack

| Layer | Technology |
|------|------------|
| Frontend | Vue 3 + Vite + Vue Router + Vue Flow |
| Backend | Spring Boot 3 + Java 17 + MyBatis-Plus |
| Database | MySQL 8 |
| Vector Retrieval | Embedding API (BAAI/bge-large-zh-v1.5) + cosine similarity |
| LLM | DeepSeek API |
| MCP | Custom SSE client + Python Flask MCP Server |
| Deployment | Docker Compose + Nginx |

### Directory Structure

```text
Agent_Weaver/
  Backend/          Spring Boot backend
  Frontend/         Vue 3 frontend
  Database/         DB initialization, seed data, snapshot support
  mcp-server/       MCP Stock service (Flask + SSE)
  deploy/           Production deployment configs
  docker-compose.local.yml
  docker-compose.prod.yml
```

### Quick Start

#### Local Docker

```bash
docker-compose -f docker-compose.local.yml up -d --build
```

Local service endpoints:

- Frontend: `http://localhost:5173`
- Backend: `http://localhost:8081`
- phpMyAdmin: `http://localhost:8080`
- MySQL: `localhost:3306`
- MCP Stock: `http://localhost:3001`

#### Production

```bash
cp .env.prod.example .env.prod
docker-compose -f docker-compose.prod.yml up -d --build
```

The production setup uses an Nginx Gateway to proxy `/api/` to the backend and `/` to the frontend. For a detailed Linux deployment walkthrough, see [云主机部署说明.md](./云主机部署说明.md).

### Database Initialization

MySQL initializes in the following priority order:

1. If an SQL snapshot exists in `Database/local-snapshot/`, it is restored first.
2. Otherwise, execute `Database/init/00-create-database.sql` → `01-schema.sql` → `seed/*.sql`.

### Notes

- `.env`, `.env.prod`, and local snapshots are excluded from Git by default.
- Replace all default passwords, JWT secrets, and encryption keys before production use.
- You may start with template values for the first deployment verification, then gradually switch to production-grade credentials.

### License

This project is licensed under the [MIT License](./LICENSE).