How to Intermediate · 3 min read

How to sandbox computer use agents

Q: How to sandbox computer use agents

Use the OpenAI Python SDK with the tools parameter specifying a computer_20241022 tool and enable the computer-use-2024-10-22 beta flag to sandbox computer use agents securely. This setup restricts the agent's environment and capabilities for safe execution.

Quick answer

Use the OpenAI Python SDK with the tools parameter specifying a computer_20241022 tool and enable the computer-use-2024-10-22 beta flag to sandbox computer use agents securely. This setup restricts the agent's environment and capabilities for safe execution.

PREREQUISITES

Python 3.8+
OpenAI API key
pip install openai>=1.0

Setup

Install the latest openai Python SDK and set your API key as an environment variable.

bash

pip install --upgrade openai

output

Collecting openai
  Downloading openai-1.x.x-py3-none-any.whl (xx kB)
Installing collected packages: openai
Successfully installed openai-1.x.x

Step by step

This example demonstrates how to create a sandboxed computer use agent using the OpenAI SDK. It configures the tools parameter with the computer_20241022 tool and enables the required beta flag.

python

import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

response = client.chat.completions.create(
    model="gpt-4o-mini",
    tools=[{
        "type": "computer_20241022",
        "name": "computer",
        "display_width_px": 1024,
        "display_height_px": 768
    }],
    betas=["computer-use-2024-10-22"],
    messages=[{"role": "user", "content": "Run a safe Python script that lists files in the sandbox."}]
)

print(response.choices[0].message.content)

output

Sandboxed file list:
- README.md
- main.py
- data.csv

Common variations

Use gpt-4o or other supported models for more capability.
Switch to async calls with asyncio and await for concurrency.
Adjust display_width_px and display_height_px to simulate different screen sizes.

python

import asyncio
from openai import OpenAI

async def main():
    client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
    response = await client.chat.completions.acreate(
        model="gpt-4o",
        tools=[{
            "type": "computer_20241022",
            "name": "computer",
            "display_width_px": 1280,
            "display_height_px": 720
        }],
        betas=["computer-use-2024-10-22"],
        messages=[{"role": "user", "content": "Check system info in sandbox."}]
    )
    print(response.choices[0].message.content)

asyncio.run(main())

output

Sandboxed system info:
OS: Linux
CPU: 4 cores
Memory: 8GB

Troubleshooting

If you get an error about missing beta flags, ensure betas=["computer-use-2024-10-22"] is included in the request.
Verify your API key has access to the computer use beta features.
Check that the tools parameter uses the exact computer_20241022 type; other values will fail.

✅

Key Takeaways

Use the OpenAI SDK's tools parameter with type 'computer_20241022' to sandbox computer use agents.
Always include the beta flag 'computer-use-2024-10-22' to enable sandboxed computer capabilities.
Adjust display dimensions to simulate different sandbox environments.
Use async calls for scalable and concurrent sandboxed agent interactions.
Check API key permissions and beta flag inclusion if sandbox features fail.

Verified 2026-04 · gpt-4o-mini, gpt-4o

Verify ↗