concave-sandbox 0.7.0

Python SDK for Concave - Secure Sandboxes for AI Agents

What's this?

This is a Python SDK for creating and managing sandboxes. These sandboxes run at scale on our infrastructure, while you can focus on using them to do anything you want.

Well, what can I do with it?

Run untrusted AI generated code, power deep research systems, environment for coding agents, train RL agents, malware analysis, or build interactive compute experiences—all in secure, high-performance sandboxes.

Features

• Secure Isolation: Complete VM-level isolation using Firecracker microVMs—every sandbox runs in its own kernel (unlike Docker containers that share the host kernel)
• Code Execution: Run Python and JavaScript code securely in isolated sandboxes
• Blazing Fast: Full VM boot up in under 200ms
• Simple API: Clean, intuitive interface with easy-to-use client SDKs
• Production Ready: Comprehensive error handling and type hints

Installation

pip install concave-sandbox

Quick Start

Configuration

Set your API key as an environment variable:

export CONCAVE_SANDBOX_API_KEY="your_api_key_here"

Sign up at concave.ai to get your API key.

Run Code

Execute Python or JavaScript code securely in isolated sandboxes:

from concave import sandbox

with sandbox() as sbx:
    result = sbx.run("print(668.5 * 2)")
    print(result.stdout) 
    
# Output: 1337.0

Manual Cleanup

If you prefer to manage the sandbox lifecycle yourself:

from concave import Sandbox

sbx = Sandbox.create()

# Execute shell commands
result = sbx.execute("uname -a")
print(result.stdout)  # Linux ...

# Run Python code
result = sbx.run("print('Hello from Python!')")
print(result.stdout)  # Hello from Python!

# Run JavaScript code
result = sbx.run("console.log('Hello from Node.js!')", language="javascript")
print(result.stdout)  # Hello from Node.js!

# Clean up
sbx.delete()

Documentation

For complete API reference, advanced examples, error handling, and best practices, visit docs.concave.ai.