This is a proof of concept for a multi agent architecture. It simulates a developer(gpt-3.5-turbo) and code reviewer(gpt-3.5-turbo-0125) working together. The input box allows you to input code into the textbox. The developer will write the code and the code reviewer will suggest improvements and the iteration will continue.
Requires Python, preferably 3.10 or higher
- Create a python virtual environment with conda or venv (optional)
conda create -n multi-agent -y python=3.10 pip
conda activate multi-agent
- Add your openai key to the streamlit secrets
Create the .streamlit folder and a file called secrets.toml, and add the key
mkdir .streamlit
touch .streamlit/secrets.toml
In the secrets.toml file put a line
OPENAI_API_KEY = "api key goes here"
- Install Requirements
pip install -r requirements.txt
- Run it
With logging to a file
streamlit run multi_agent.py --logger.level=info 2>'app.log'
or log to terminal
streamlit run multi_agent.py