cillow 1.0.0

Cillow is an open-source library that enables you to execute AI-generated code in a controlled environment. The name "Cillow" follows the same naming convention as "Pillow" (Python Imaging Library), representing its role as a Code Interpreter Library.

It offers key features such as:

• Environment Switching: Effortlessly switch between multiple Python environments.
• Automated Package Installation: Automatically install imported packages using uv or pip.
• Functionality Patches: Apply patches to restrict the scope of AI-generated code, capture outputs such as stdout, stderr, images, plots, and more.

Check Documentation

Visit synacktra.is-a.dev/cillow

Installation

pip install cillow

Hosting a server

import cillow

cillow.add_patches(
    cillow.prebuilt_patches.patch_stdout_stderr_write,
    cillow.prebuilt_patches.patch_matplotlib_pyplot_show,
    cillow.prebuilt_patches.patch_pillow_show,
)

if __name__ == "__main__":
    server = cillow.Server(
        port=5556, max_interpreters=2, interpreters_per_client=1
    )
    server.run()

Running code through a client

import cillow

client = cillow.Client.new(host="127.0.0.1", port=5556)

client.run_code("""
from PIL import Image, ImageDraw

img = Image.new('RGB', (400, 300), 'white')

draw = ImageDraw.Draw(img)
draw.rectangle([50, 50, 350, 250], outline='black', width=3)
draw.ellipse([100, 100, 300, 200], outline='purple', width=3)
draw.line([50, 250, 350, 250], fill='blue', width=3)

img.show()
""")

At the moment, Cillow only supports Python, as it does not rely on Jupyter Kernel/Lab.

This project began as an exploration of E2B's code interpreter. I implemented the Python interpreter from scratch, taking a different approach by adding features like environment switching and functionality patching. Recognizing the potential of the project, I expanded it into a client-server architecture using ZeroMQ, threading, and multiprocessing.