Agent Zero

Agent Zero is a project that i made for myself so as to improve the overall quality of the agentic system. This project demonstrates the use of multiple AI agents working together in a group chat environment. It utilizes OpenAI's GPT 4o model and Anthropic's Claude 3.5 sonnet model to create a collaborative AI system.

Features

• Multiple AI agents: OpenAI assistant, Anthropic assistant (Claude), and a dedicated code writer agent (Currently configured to use claude)
• Group chat management for coordinating agent interactions
• Exponential backoff retry mechanism for handling rate limits
• YAML-based prompt configuration

Note: This execution asks for user input after each assistant speaks, you can turn this off by changing human_input = "NEVER"

Prerequisites

• Python 3.7+
• OpenAI API key
• Anthropic API key

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/agent-zero.git
   cd agent-zero
   

2. Install the required dependencies:

   pip install -r requirements.txt
   

3. Set up your environment variables:

   export OPENAI_API_KEY='your_openai_api_key_here'
   export ANTHROPIC_API_KEY='your_anthropic_api_key_here'
   

Usage

1. Prepare your prompt in a prompt.yaml file:

   prompt: "Your prompt text here"

2. Run the main script:

   python agent_zero.py
   

Configuration

• Modify the openai_llm_config and claude_llm_config dictionaries in agent_zero.py to adjust model parameters or API endpoints.
• The GroupChat is configured for a maximum of 12 rounds. Adjust the max_round parameter in the GroupChat initialization if needed.

Troubleshooting

• If you encounter rate limit errors, the script will automatically retry with exponential backoff.
• Ensure that your API keys are correctly set in the environment variables.
• Check that the prompt.yaml file is in the same directory as the script and contains a valid prompt.

Contributing

Contributions to Agent Zero are welcome! Please feel free to submit a Pull Request.