new generation from 15/03/2025,
mainly based on stable version of LangGraph & LangChain, ensembling MCP from Anthropic and Browser use.
more info refer to new project repo: AdGraph
Multi-Agent System Workflow for Software Engineering
In software development scenarios, multiple roles—such as project managers, tech leads, and QA engineers—collaborate to deliver a project demo that meets customer requirements. By building role-specific agents powered by domain expertise, we can significantly reduce development costs and streamline the process. To this end, we propose MAS4SE (Multi-Agent System for Software Engineering) , a platform built upon the open-source LLM framework LangChain and the multi-agent orchestration framework LangGraph .
- Historical cases showcase
- Agent customized by User, Workflow UI like Coze
- Chat with Service Bot
- Collaboration process of Agents
Please ensure you have the following environments :
- Python 3.9 + LangChain + LangGraph
- Node.js + Vue3
- Ollama
- Redis
-
Clone this repo
git clone https://github.com/waywooKwong/CSI-LangChain-LLM-Chatbot.git -
run Web UI
cd MAS4SE-frontend npm run serve -
run backend function server
cd MAS4SE-backend python run.py -
run database server
cd MAS4SE-frontend/src/database cd Redis redis-server.exe node server.cjs
Attention：
- embedding_models: m3e-base needs to be downloaded local.
MAS4SE-backend/src/embedding_models/m3e-base
- Interaction between frontend & backend
前端 JointUI 向后端传递连接结点关系
MAS4SE-backend/frontend_json_process/json_simplified
工作流运行完成后生成的文本
MessageSum
Origin version: Weihua Kuang (邝伟华)、Pu Wang (王璞)、Cheng Qian (钱程)、Wenxiang Yu (余文祥)
This repository is maintained by Weihua Kuang (@waywooKwong).
图片中展示的是精简后只保留核心功能代码的工作量，但要注意，仓库中保存的历史v1版本是完整原始开发的代码。 包含冒烟测试代码以及预学习的模型组件代码，实际的工作量（指Python）部分应当是当前的 2-3 倍。
Prebuid-version 的工作量
We use JointJS+ to implement the interactive workflow part of Web UI.
https://www.jointjs.com/demos/chatbot
Push issues under this repo, or eamil Weihua Kuang by weihua.kwong@mail.nankai.edu.cn