OpenSandbox  by alibaba

Sandbox platform for AI and LLM applications

Created 1 month ago
560 stars

Top 57.4% on SourcePulse

GitHubView on GitHub
Project Summary

OpenSandbox is a universal sandbox platform designed for AI application scenarios, offering multi-language SDKs, a unified sandbox protocol, and flexible runtimes for LLM-related capabilities. It targets engineers and researchers needing to standardize interactions with diverse execution environments, such as command execution, file operations, code interpretation, and agent execution, thereby accelerating AI development and deployment.

How It Works

The platform defines a unified sandbox protocol for managing sandbox lifecycles and execution APIs, allowing for extensibility. It provides SDKs in Python, Java/Kotlin, and JavaScript/TypeScript, with Go support on the roadmap. The system supports Docker and Kubernetes runtimes for scalable, distributed sandbox scheduling. Built-in environments include Command, Filesystem, and Code Interpreter, with examples for integrating Coding Agents, browser automation, and desktop environments. This approach offers a consistent interface across varied execution backends.

Quick Start & Requirements

  1. Clone the repository: git clone https://github.com/alibaba/OpenSandbox.git && cd OpenSandbox
  2. Start the sandbox server: Navigate to server/, copy example.config.toml to ~/.sandbox.toml, and run uv run python -m src.main.
  3. Install the Code Interpreter SDK: uv pip install opensandbox-code-interpreter.
  4. Execute provided Python examples.
  • Prerequisites: Docker (required for local execution), Python 3.10+ (recommended).
  • Links: Examples are located in the examples/ directory. Architecture details are in docs/architecture.md.

Highlighted Details

  • Multi-language SDKs: Python, Java/Kotlin, JavaScript/TypeScript available, with Go on the roadmap.
  • Flexible Runtimes: Supports Docker and Kubernetes for distributed sandbox scheduling.
  • Diverse Environments: Includes built-in Command, Filesystem, and Code Interpreter, with examples for Coding Agents (Claude, Gemini, Codex), Browser automation (Playwright), and Desktop environments (VNC, VS Code Web).
  • Unified Protocol: Standardizes sandbox lifecycle and execution APIs for interoperability.

Maintenance & Community

The project is actively developed, with GitHub Actions for continuous integration. Community discussions and feature requests are managed via GitHub Issues.

Licensing & Compatibility

Licensed under the Apache 2.0 License, permitting use for personal or commercial projects in compliance with license terms.

Limitations & Caveats

Go SDK support is currently on the roadmap. Kubernetes runtime implementation is ongoing. Advanced network egress control features, such as declarative isolation and Layer 2 network control, are also listed as future roadmap items.

Health Check
Last Commit

1 day ago

Responsiveness

Inactive

Pull Requests (30d)
63
Issues (30d)
14
Star History
503 stars in the last 30 days

Explore Similar Projects

Starred by Wes McKinney Wes McKinney(Author of Pandas), Chip Huyen Chip Huyen(Author of "AI Engineering", "Designing Machine Learning Systems"), and
22 more.

autogen by microsoft

0.5%
54k
Agentic framework for multi-agent AI applications
Created 2 years ago
Updated 5 days ago
Feedback? Help us improve.