Agent Sandbox is a next-generation runtime infrastructure specifically designed for AI Agents, providing a secure and isolated cloud sandbox environment for executing AI-generated code.

​
Features

  • Secure. Sandboxes have system-level isolation capabilities, ensuring the security of data and execution environment within the sandbox.
  • Fast startup. Startup time is less than 200ms, with support for large-scale concurrent creation, suitable for AI application scenarios with high concurrency requirements.
  • Multi-language execution support. Supports running code in multiple programming languages including Python, JavaScript, TypeScript, C++, and more.
  • Quick pause and resume. Can pause the sandbox at any time and resume it when needed later, with filesystem and process state also being restored.
  • Background execution. Supports background task execution, suitable for scenarios that require waiting for results.
  • Easy integration. Seamlessly integrates with popular AI Agent frameworks and tools.

​
Scenarios

  • AI data analysis and visualization;
  • Code execution environment for Code Agents;
  • Computer Use Agent.

Or contact our solution engineers to learn how to integrate into your business.

​
Pricing

Agent Sandbox supports billing for CPU and RAM (memory) by the second, with on-demand usage and flexible billing. Storage and templates are currently free. For details, please refer to pricing page.

​
Getting Started

You can start and manage your Agent Sandboxes through SDK or CLI. For details, please refer to quick start page.