OpenSandbox

✨Key Features

• ▸Multi-Language SDKs: Official support for Python, Java/Kotlin, JavaScript/TypeScript, C#/.NET (Go in roadmap), enabling easy integration from various AI/agent frameworks.
• ▸Unified Sandbox APIs: Standardized lifecycle (create/run/destroy) and execution APIs for commands, file operations, code interpretation, resource monitoring, and custom extensions.
• ▸Runtimes & Deployment: Built-in Docker for local/quick testing; Kubernetes for scalable, distributed production clusters; supports secure isolation layers like gVisor, Kata Containers, and Firecracker microVMs.
• ▸Pre-built Environments: Includes Command sandbox, Filesystem access, Code Interpreter (e.g., safe Python execution), browser automation (Chrome/Playwright), desktop environments (VNC/VS Code), and more.
• ▸Network & Security Controls: Ingress Gateway for routing, per-sandbox egress policies, strong isolation to prevent escapes or network abuse when running LLM-generated or agent code.
• ▸Use Cases & Integrations: Designed for autonomous AI agents (e.g., Claude Code, Gemini CLI, OpenAI Codex, LangGraph), agent evaluation sandboxes, RL training loops, browser/desktop automation, and safe execution of dynamic/untrusted code.
• ▸Open-Source & Active: Apache 2.0 license; ~7.3k stars, 541 forks, 34 contributors, 53 releases, very active maintenance (commits as recent as March 10, 2026).
• ▸Documentation & Getting Started: Full docs at open-sandbox.ai (English/Chinese), examples for agents/RL/browser, easy local setup via pip install opensandbox-server + config.