AgentCoder is a multi-agent framework for generating high-quality code using LLMs. It targets developers and researchers seeking to improve code generation through specialized, collaborating agents and an iterative feedback loop for refinement.

How It Works

AgentCoder employs three specialized agents: a programmer agent for code generation, a test designer agent for independent test case creation, and a test executor agent to run tests and provide feedback. This multi-agent approach allows for task specialization and an iterative refinement process, aiming for more robust and correct code output.

Quick Start & Requirements

• Install dependencies: pip install -r requirements.txt
• Set OpenAI API key in .env file.
• Usage examples:
  • Code Generation: python programmer_[humaneval/mbpp].py
  • Test Case Generation: python test_designer_[humaneval/mbpp].py
  • Self-Optimization: python test_executor_[humaneval/mbpp].py
• Requires an OpenAI API key or equivalent.

Highlighted Details

• Multiagent collaboration with specialized roles.
• Independent test case generation by a dedicated agent.
• Iterative code refinement based on test execution feedback.
• Modular design for integration with advanced models.

Maintenance & Community

Contributions are welcome via issues or pull requests. The project acknowledges funding and support from AIOHUB.

Licensing & Compatibility

Released under the MIT License, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

The framework relies on external LLM APIs, and its performance is dependent on the quality of these models and the provided API keys.