Running untrusted or AI-generated code can compromise local environments or lead to unintended side effects. LitSandbox provides a secure, disposable, and isolated environment specifically designed for executing AI generated code safely.

Experiment safely and build prototypes without compromising security.

✅ Run code in a sandbox     ✅ Integrate with MCP Server  ✅ Support GPU machine
✅ No MLOps glue code        ✅ Easy setup in Python       ✅ Install any package

Lightning AI • Docs • Quick start

Quick start

Install LitSandbox via pip:

pip install litsandbox

Example

from litsandbox import Sandbox

s = Sandbox()
output = s.run('git clone https://github.com/octocat/Hello-World.git')
s.run_python_code("f = open('file.txt', 'w'); f.write('hello world')")
s.stop()

Key benefits

• Run code in a safe, isolated environment
• Data persists across restarts
• Supports GPU machines
• Integrate with MCP Server
• Easy setup in Python
• Install any package

Build agentic system

Safely prototype with LLM-generated code using a sandboxed execution environment.

from openai import OpenAI
from litsandbox import Sandbox


# Prompt the model to generate Python code
client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4.1",
    messages=[
        {"role": "system", "content": "You are a helpful Python coding assistant. Only return clean, executable Python code with no explanations. Your response will be executed directly."},
        {"role": "user", "content": "Write a Python program that prints the first 5 square numbers."}
])
generated_code = response.choices[0].message.content.strip()
print("Generated Code:\n", generated_code)


# Execute the code securely in a sandbox
sandbox = Sandbox(machine="CPU")
output = sandbox.run_python_code(generated_code)
print("\nExecution Output:\n", output.text)
# Stop the sandbox
sandbox.stop()