MetaGPT的实际使用结果

MetaGPT就是输入一句话的老板需求，输出用户故事 / 竞品分析 / 需求 / 数据结构 / APIs / 文件等。看着非常美好，但是实际情况怎么样呢，网上搜了搜，全是垃圾信息，只好花钱自己试试了，本文就是对其结果的记录。

1 生成个项目

我这里是使用docker

• docker的映射文件夹

  mkdir -p /opt/metagpt/{config,workspace}

• 建立配置文件

  vim /opt/metagpt/config.yaml
  
  填入：
    OPENAI_API_KEY: "sk-m你的gpt key"
    OPENAI_API_BASE: "https://你的openai代理/v1"
    OPENAI_API_MODEL: "gpt-4"
  
  cp /opt/metagpt/config/config.yaml /opt/metagpt/config/key.yaml

• 拉取容器

  docker pull metagpt/metagpt:v0.3

• 启动容器

  docker run --name metagpt -d \
  --privileged \
  -v /opt/metagpt/config:/app/metagpt/config \
  -v /opt/metagpt/workspace:/app/metagpt/workspace \
  metagpt/metagpt:v0.3

• 进入容器生成项目

  docker exec -it metagpt /bin/bash
  python startup.py "使用python写一个API网关，支持负载均衡，拥有web管理平 台"

  首次运行时报错：

  Traceback (most recent call last):
  File "/app/metagpt/startup.py", line 29, in <module>
      fire.Fire(main)
  File "/usr/local/lib/python3.9/site-packages/fire/core.py", line 141, in Fire
      component_trace = _Fire(component, args, parsed_flag_args, context, name)
  File "/usr/local/lib/python3.9/site-packages/fire/core.py", line 466, in _Fire
      component, remaining_args = _CallAndUpdateTrace(
  File "/usr/local/lib/python3.9/site-packages/fire/core.py", line 681, in _CallAndUpdateTrace
      component = fn(*varargs, **kwargs)
  File "/app/metagpt/startup.py", line 25, in main
      asyncio.run(startup(idea, investment, n_round))
  File "/usr/local/lib/python3.9/asyncio/runners.py", line 44, in run
      return loop.run_until_complete(main)
  File "/usr/local/lib/python3.9/asyncio/base_events.py", line 647, in run_until_complete
      return future.result()
  File "/app/metagpt/startup.py", line 15, in startup
      await company.run(n_round=n_round)
  File "/app/metagpt/metagpt/software_company.py", line 60, in run
      await self.environment.run()
  File "/app/metagpt/metagpt/environment.py", line 64, in run
      await asyncio.gather(*futures)
  File "/app/metagpt/metagpt/roles/role.py", line 229, in run
      rsp = await self._react()
  File "/app/metagpt/metagpt/roles/role.py", line 200, in _react
      return await self._act()
  File "/app/metagpt/metagpt/roles/engineer.py", line 143, in _act
      return await self._act_sp()
  File "/app/metagpt/metagpt/roles/engineer.py", line 128, in _act_sp
      code_rsp = await WriteCode().run(
  File "/app/metagpt/metagpt/actions/write_code.py", line 71, in run
      self._save(context, filename, code_rsp)
  File "/app/metagpt/metagpt/actions/write_code.py", line 60, in _save
      code = CodeParser.parse_code(block="", text=code_rsp)
  File "/app/metagpt/metagpt/utils/common.py", line 172, in parse_code
      raise Exception
  Exception

  异常居然没有写个原因，我也懒得去看具体原因，盲猜流式接收接口应答途中挂了。再次运行成功结束。总共用了18分钟，花费1.3美元

2 生成结果

生成文件在workspace中

├── [4.0K]  api_gateway
│   ├── [ 771]  admin.py
│   ├── [ 557]  load_balancer.py
│   ├── [1.4K]  main.py
│   ├── [ 598]  models.py
│   └── [1.4K]  views.py
├── [4.0K]  docs
│   ├── [2.6K]  prd.md
│   └── [1.5K]  system_design.md
└── [4.0K]  resources
    ├── [ 425]  competitive_analysis.mmd
    ├── [ 17K]  competitive_analysis.pdf
    ├── [ 32K]  competitive_analysis.png
    ├── [5.1K]  competitive_analysis.svg
    ├── [ 475]  data_api_design.mmd
    ├── [ 20K]  data_api_design.pdf
    ├── [ 34K]  data_api_design.png
    ├── [ 12K]  data_api_design.svg
    ├── [ 319]  seq_flow.mmd
    ├── [ 15K]  seq_flow.pdf
    ├── [ 32K]  seq_flow.png
    └── [ 22K]  seq_flow.svg

其中文档prd.md中包含最开始我提供的需求扩展，比如什么产品目标、竞品分析、需求分析、需求池什么的，看着像那么回事儿，并且内容符合我的初衷，但是肯定是没法和正式这类文档比较，属于大学课程设计水平吧。

文档system_design.md描述了数据结构和程序流程，从里面非常简单的表述我感到不妙。

3 运行

直接执行python main.py，果不其然，报错了，找不到Base。下列是main.py函数

## Required Python third-party packages
from flask import Flask
from sqlalchemy import create_engine
from sqlalchemy.orm import sessionmaker
from models import API
from load_balancer import LoadBalancer
from views import APIView
from admin import AdminInterface
from flask_admin.contrib.sqla import ModelView

## Flask application setup
app = Flask(__name__)

## Database setup
engine = create_engine('sqlite:///api_gateway.db')
Session = sessionmaker(bind=engine)

## Create tables
Base.metadata.create_all(engine)

## Create a session
session = Session()

## Create API instances
api1 = API(name='API1', endpoint='http://localhost:5000/api1', method='GET', is_active=True)
api2 = API(name='API2', endpoint='http://localhost:5000/api2', method='GET', is_active=True)

## Add API instances to the session
session.add(api1)
session.add(api2)
session.commit()

## Create LoadBalancer instance
load_balancer = LoadBalancer([api1, api2])

## Create AdminInterface instance
admin_interface = AdminInterface(app, session)

## Add API view to the admin interface
admin_interface.add_view(ModelView(API, session))

## Create APIView instance
api_view = APIView(session)

## Add API routes to the Flask application
app.add_url_rule('/api/<api_name>', view_func=api_view, methods=['GET', 'DELETE'])
app.add_url_rule('/api', view_func=api_view, methods=['POST'])

## Run the Flask application
if __name__ == '__main__':
    app.run(debug=True)

很明显在使用前没定义，找了找在models.py中定义了Base，但是没有导入，于是添加后继续运行，继续报错，这次是在views.py中

## Required Python third-party packages
from flask import Flask, request, jsonify
from flask.views import MethodView
from sqlalchemy.orm import Session

## Data structures and interface definitions
class APIView(MethodView):
    def __init__(self, session: Session):
        self.session = session

    def get(self, api_name: str):
        """
        This method returns the details of an API.
        """
        api = self.session.query(API).filter_by(name=api_name).first()
        if not api:
            return jsonify({'error': 'API not found'}), 404
        return jsonify(api.__dict__), 200

    def post(self):
        """
        This method adds a new API.
        """
        data = request.get_json()
        api = API(**data)
        self.session.add(api)
        self.session.commit()
        return jsonify(api.__dict__), 201

    def delete(self, api_name: str):
        """
        This method deletes an API.
        """
        api = self.session.query(API).filter_by(name=api_name).first()
        if not api:
            return jsonify({'error': 'API not found'}), 404
        self.session.delete(api)
        self.session.commit()
        return jsonify({'success': 'API deleted'}), 200

## Flask application setup
app = Flask(__name__)
api_view = APIView(Session())

app.add_url_rule('/api/<api_name>', view_func=api_view, methods=['GET', 'DELETE'])
app.add_url_rule('/api', view_func=api_view, methods=['POST'])

报错说app.add_url_rule('/api/<api_name>', view_func=api_view, methods=['GET', 'DELETE'])中的api_view没有__name__属性。先不管错误，最后这四行全局运行的和主函数里面一摸一样，这明显结构有问题。懒得再继续改了，这些生成的文件最多也就40多行，实现了些及其简单的功能，连个简单架子都算不上，应该是没有任何用处的垃圾。

4 结论

平日里也经常用gpt生成代码来用，但也只能使其生成最末端的功能，该开源仓库的思路其实挺棒的，但就目前最优gpt4方案仍然达不到自我细分后，实现组装的能力，所有目前来看它不能实际用来做什么，更多的是提供一种研究思路。