Getting Started with OpenSandbox: A Step-by-Step Tutorial

Getting Started with OpenSandbox: A Practical Tutorial

In this tutorial, we'll walk through installing OpenSandbox, creating your first sandbox, running code in it, and integrating it with AI agents. By the end, you'll have a fully functional sandbox environment for your AI workloads.

Step 1: Installation via Docker

The quickest way to get started with OpenSandbox is using Docker. Pull the official image and start a sandbox server:

docker pull opensandbox/execd:latest
docker run -d --name opensandbox \
  -p 8080:8080 \
  -v ./data/opensandbox:/data \
  opensandbox/execd:latest

This starts the sandbox runtime on port 8080. You can verify it's running with docker logs opensandbox.

Step 2: Python SDK Quick Start

Install the Python SDK and create your first sandbox:

pip install opensandbox

from opensandbox import SandboxClient

# Connect to your sandbox server
client = SandboxClient("http://localhost:8080")

# Create a sandbox
sandbox = client.create_sandbox(
    image="python:3.11",
    resources={"cpu": 1, "memory_mb": 512}
)

# Execute code
result = sandbox.run("print('Hello from OpenSandbox!')")
print(result.output)  # Hello from OpenSandbox!

# Clean up
sandbox.destroy()

Step 3: Running Code Safely

OpenSandbox shines when you need to execute untrusted code safely. Here's how to use it with a coding agent:

# Execute arbitrary code with resource limits
result = sandbox.run(
    code="""
    import subprocess
    result = subprocess.run(['ls', '-la'], capture_output=True, text=True)
    print(f"Files: {result.stdout}")
    """,
    timeout=30,  # 30-second timeout
    memory_limit_mb=256
)
print(result.exit_code)  # Safe execution

Step 4: Kubernetes Deployment

For production deployments, OpenSandbox supports Kubernetes with automatic pod scheduling, resource management, and horizontal scaling. The CNCF Landscape-listed project provides Helm charts for easy deployment:

helm repo add opensandbox https://opensandbox-group.github.io/helm-charts
helm install opensandbox opensandbox/opensandbox \
  --set replicas=3 \
  --set runtime.engine=kubernetes

Step 5: Integrating with MCP

OpenSandbox provides an MCP server that integrates seamlessly with AI agent frameworks:

# Configure your AI agent to use OpenSandbox MCP
mcp_servers:
  opensandbox:
    command: npx @opensandbox/mcp-server
    args:
      - --endpoint=http://localhost:8080

Comparison: OpenSandbox vs Other Solutions

Conclusion

OpenSandbox is the most complete sandbox platform for AI agents available today. With 12K+ GitHub stars, CNCF landscape listing, and a vibrant community, it's proven in production at scale. The multi-language SDKs, Kubernetes support, and strong isolation features make it an essential tool for any serious AI development workflow.