Pythoneer AI is a tool to help developers make codebase-wide changes to their Python projects - giving them more time to focus on more enjoyable and rewarding problems.

The agent is provided with a codebase and a task, and takes iterative steps to complete it. Through the use of tools, the agent is able to open, edit and create files, and run tests and python scripts.

• Porting Python 2 to Python 3
• Converting PyTorch codebases to TensorFlow
• Converting TensorFlow codebases to PyTorch

Check out the website for some demos!

• Install Docker 🐋. Docker is used by the agent to run Python scripts in isolated, pre-defined environments. See Get Docker for installation instructions.
• Run docker-compose build to build the required images.
• Create a .env file at the project root, and add an entry for your ANTHROPIC_API_KEY.
• Create a virtual environment and run pip install . to install the package and the dependencies.

• Create a config file, or use one of the existing configs in /config. The config file defines the task, and the tools that the agent has access to.
• Run the agent:

export CONFIG_FILE=<path to config file>
export CODEBASE_PATH=<path to codebase to edit>
export WORKSPACE_PATH=<path to directory to save outputs to>

run.py \
--config_file $CONFIG_FILE \
--codebase_path $CODEBASE_PATH \
--workspace_path $WORKSPACE_PATH