svcvit/dify-sandbox-py

https://github.com/svcvit/dify-sandbox-py

A code executor for Dify that is compatible with the official sandbox API calls and dependency installation.

• Supports Python 3.12
• Supports Node.js 20

Purpose

While the official sandbox has many permission settings and is a better sandboxing solution, in personal use cases where Dify's code nodes are entirely self-edited, there's no risk of code injection. This project aims to provide broader permissions and support for more dependencies (like numpy>2.0, matplotlib, scikit-learn) to reduce confusing error messages. This code was developed by referencing the official sandbox's API call examples.

Usage

In the official docker-compose.yaml, locate the sandbox image section and replace it with:

  sandbox:
    # image: langgenius/dify-sandbox:0.2.10
    image: svcvit/dify-sandbox-py:0.1.1

If you prefer to build the image yourself, you can clone this repository and run:

docker build -t dify-sandbox-py:local .

Then modify the sandbox image in docker-compose.yaml to use dify-sandbox-py:local

Notes

• Network access control has been removed, network access is enabled by default
• Third-party dependency installation works the same as the official version - place your dependencies in /docker/volumes/sandbox/dependencies/python-requirements.txt and restart the sandbox
• The image only includes FastAPI-related dependencies by default. Any additional dependencies you need must be added to python-requirements.txt