Multi-agent framework, the first AI software company, enabling natural language programming

MetaGPT - Detailed Introduction to the Multi-Agent Framework Project

Project Overview

MetaGPT is an open-source multi-agent framework that coordinates human programming knowledge and AI agents powered by large language models (LLMs) to develop diverse software solutions. The core concept of the project is Code = SOP(Team), which materializes Standard Operating Procedures (SOPs) and applies them to a team composed of LLMs.

Project Slogan

🌟 Multi-Agent Framework: The First AI Software Company, Moving Towards Natural Language Programming

Core Features

1. Generate a Complete Project from a Single Line of Requirements

• Input: A single line of requirement description
• Output: User stories, competitive analysis, requirement documents, data structures, APIs, technical documentation, etc.

2. Simulate a Complete Software Company Process

MetaGPT internally contains AI agents with different roles:

• Product Manager - Responsible for requirements analysis and product planning
• Architect - Responsible for system design and technical architecture
• Project Manager - Responsible for project management and progress control
• Engineer - Responsible for code implementation and development

3. Standardized Operating Procedures

Ensures the standardization and efficiency of the entire software development process through carefully orchestrated SOPs (Standard Operating Procedures).

Latest Developments

Important Milestones in 2025

• March 10th: mgx.dev became #1 Product of the Week on Product Hunt
• March 4th: mgx.dev became #1 Product of the Day on Product Hunt
• February 19th: Officially released the natural language programming product MGX (MetaGPT X) - the world's first AI agent development team
• February 17th: Released two papers: SPO and AOT
• January 22nd: The paper "AFlow: Automating Agentic Workflow Generation" was accepted as an oral presentation (top 1.8%) at ICLR 2025, ranking 2nd in the LLM-based Agent category

Technical Requirements

System Requirements

• Python 3.9 or higher, but lower than 3.12
• Node.js and pnpm (for actual use)

Environment Check

python --version

Create Environment

conda create -n metagpt python=3.9 && conda activate metagpt

Installation Methods

Method 1: pip Installation

pip install --upgrade metagpt

Method 2: Install from GitHub

pip install --upgrade git+https://github.com/geekan/MetaGPT.git

Method 3: Source Code Installation

git clone https://github.com/geekan/MetaGPT && cd MetaGPT && pip install --upgrade -e .

Configuration Settings

Initialize Configuration

metagpt --init-config  # Create ~/.metagpt/config2.yaml

Configuration File Example

llm:
  api_type: "openai"  # Or azure / ollama / groq etc.
  model: "gpt-4-turbo"  # Or gpt-3.5-turbo
  base_url: "https://api.openai.com/v1"
  api_key: "YOUR_API_KEY"

Usage Methods

Command Line Usage

metagpt "Create a 2048 game"  # Create a code repository in ./workspace

Using as a Library

from metagpt.software_company import generate_repo
from metagpt.utils.project_repo import ProjectRepo

repo: ProjectRepo = generate_repo("Create a 2048 game")
print(repo)  # Print the code repository structure containing files

Data Interpreter Usage

import asyncio
from metagpt.roles.di.data_interpreter import DataInterpreter

async def main():
    di = DataInterpreter()
    await di.run("Run data analysis on sklearn Iris dataset, include a plot")

asyncio.run(main())  # Or use await main() in jupyter notebook

Online Experience

• Hugging Face Space: MetaGPT Software Company Space
• Video Tutorial: Installation tutorial by Matthew Berman
• Official Demo: Complete demo video available on the project's GitHub page

Resource Links

Documentation and Guides

• Online Documentation: docs.deepwisdom.ai
• Usage Guide: Quick start and detailed usage instructions
• Feature Introduction: What MetaGPT can do
• Building Guide: How to build your own agent
• Frequently Asked Questions: FAQ answers

Community and Support

• Discord Channel: Join the community discussion
• GitHub Issues: Technical issues and feedback
• Twitter: @MetaGPT_ Get the latest updates
• Email Contact: alexanderwu@deepwisdom.ai

Summary

MetaGPT represents a significant breakthrough in AI-assisted software development, achieving fully automated generation from requirements to code through multi-agent collaboration and standardized processes. It is not only a technical tool but also an exploration and practice of future software development models, providing a feasible solution for the realization of natural language programming.