Python Exec Sandbox

About

This tool offers a robust solution for AI agents requiring Python execution, preventing host system pollution and dependency conflicts. It creates ephemeral, isolated environments using bubblewrap on Linux or Docker on macOS, where scripts run with inline PEP 723 dependencies and `uv` for efficient package resolution and multi-version Python support. Each execution is clean, isolated, and disposable, ensuring a pristine host environment free from package accumulation, virtual environment sprawl, or rogue `pip install` issues.

Key Features

• PEP 723 inline metadata for declaring script dependencies directly
• Multi-version Python support, automatically downloading Python 3.13, 3.14, or 3.15 via uv
• Ephemeral environments where dependencies are resolved per-execution and never persisted
• Efficient package caching for near-instant repeat dependency installations
• Sandboxed execution with platform-specific isolation (bubblewrap on Linux, Docker on macOS)
• 0 GitHub stars

Use Cases

• Running and testing Python code across different Python versions for agent tasks
• Securely executing Python scripts from AI agents without host filesystem access
• Maintaining a clean and consistent host environment for AI agent development