AgileCoder is a multi-agent framework for automated software development, inspired by Agile methodologies. It targets developers and researchers seeking to automate complex software creation by dynamically managing tasks and sprints, aiming for more realistic and adaptable development cycles compared to static role-based agents.

How It Works

AgileCoder employs a task-oriented approach, creating a dynamic backlog that is updated across development sprints. This mimics real-world Agile practices by allowing for iterative refinement and adaptation, rather than assigning fixed roles to agents. This dynamic task management is key to its ability to handle complex software requirements.

Quick Start & Requirements

• Install via pip: pip install agilecoder
• Supports OpenAI, Azure OpenAI, Anthropic, and Ollama self-hosted models.
• Configuration requires setting API keys and endpoints via environment variables or a .env file.
• Usage: agilecoder --task "Description of software to create"
• Documentation: setup_model.md

Highlighted Details

• Outperforms ChatDev and MetaGPT on HumanEval and MBPP benchmarks with Claude 3 Haiku and GPT-4.
• Achieves a 57.79% executability rate on complex software requirements (ProjectDev dataset), significantly higher than competitors.
• Reports 0 errors in ProjectDev evaluation, suggesting robust error handling or reporting.
• Demonstrates a higher token usage and cost compared to MetaGPT, but with improved executability.

Maintenance & Community

• Project is associated with FSoft-AI4Code.
• Contact email: bdqnghi@gmail.com
• Website: fpt-aicenter
• Paper available at: arXiv:2406.11912

Licensing & Compatibility

• MIT License.
• Permissive for commercial use and integration with closed-source projects.

Limitations & Caveats

The framework exhibits significantly higher token usage and associated costs compared to some alternatives like MetaGPT. The average time per sprint and total runtime are also considerably higher, indicating potential performance bottlenecks for rapid development cycles.