MetaGPT is a multi-agent framework designed to simulate an AI software company, transforming a single natural language requirement into a complete software project. It targets developers and researchers interested in natural language programming and AI-driven software development, offering a structured approach to complex task automation.

How It Works

MetaGPT operates on the philosophy of "Code = SOP(Team)," materializing Standard Operating Procedures (SOPs) and applying them to teams of Large Language Models (LLMs). It assigns distinct roles (e.g., product manager, architect, engineer) to LLMs, enabling them to collaborate and execute a software development lifecycle, from initial requirements to final code. This multi-agent, SOP-driven approach aims to orchestrate complex workflows more effectively than single-agent systems.

Quick Start & Requirements

• Installation: pip install --upgrade metagpt or pip install --upgrade git+https://github.com/geekan/MetaGPT.git. Requires Python 3.9 to < 3.12. Node.js and pnpm are needed for full functionality.
• Configuration: Initialize with metagpt --init-config to create ~/.metagpt/config2.yaml, which requires LLM API keys and model configurations.
• Usage: CLI: metagpt "Create a 2048 game". Library: from metagpt.software_company import generate_repo.
• Resources: Official documentation and a demo video are available.

Highlighted Details

• Simulates a full software company with roles like PM, architect, and engineer.
• Outputs user stories, competitive analysis, requirements, data structures, APIs, and documents.
• Supports various LLM providers (OpenAI, Azure, Ollama, Groq) via configuration.
• Includes a Data Interpreter agent for code execution and analysis.

Maintenance & Community

The project has active development with recent papers on SPO and AOT, and an accepted paper (AFlow) at ICLR 2025. A Discord channel is available for community support and contribution.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README text. Users should verify licensing for commercial or closed-source integration.

Limitations & Caveats

The framework requires specific Python versions (3.9 to < 3.12) and external dependencies like Node.js. LLM API costs are not detailed. The "Software Company Multi-Agent Schematic" is noted as "Gradually Implementing," suggesting ongoing development and potential incompleteness of certain features.