OpenSandbox is a general-purpose sandbox platform built for AI applications. It exposes a unified lifecycle API backed by Docker and Kubernetes runtimes, ships multi-language SDKs and a CLI, and integrates with MCP-capable agents. Whether you are running a coding agent, evaluating model outputs, or executing untrusted code, OpenSandbox handles sandbox creation, command execution, file operations, and network policy — all from a single server process.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/opensandbox-group/OpenSandbox/llms.txt
Use this file to discover all available pages before exploring further.
Key Features
Multi-Language SDKs
First-class SDKs for Python, JavaScript/TypeScript, Kotlin/Java, Go, and C#/.NET covering sandbox lifecycle, command execution, and file operations.
Docker & Kubernetes Runtimes
Built-in support for local Docker execution and high-performance distributed scheduling on Kubernetes — same API, different scale.
osb CLI
A full-featured terminal interface for creating sandboxes, running commands, moving files, inspecting diagnostics, and managing egress policy.
MCP Server
Exposes sandbox creation, command execution, and file operations to any MCP-capable client such as Claude Code and Cursor.
Credential Vault
Secure credential injection for sandbox outbound requests so real secrets are never exposed to the sandbox workload itself.
Secure Runtimes
Optional strong-isolation container runtimes: gVisor, Kata Containers, and Firecracker microVM for enhanced host-workload separation.
Code Interpreter
Higher-level SDKs for multi-language code execution inside sandboxes, available for Python, JavaScript/TypeScript, Kotlin/Java, and C#/.NET.
Network Policy
Unified ingress gateway with multiple routing strategies plus per-sandbox egress controls to restrict outbound traffic from individual sandboxes.
Use-Case Scenarios
OpenSandbox is designed for a wide range of AI workloads. The table below maps common scenarios to the capabilities they rely on.| Scenario | What OpenSandbox provides |
|---|---|
| Coding Agents | Isolated environments for Claude Code, Gemini CLI, OpenAI Codex CLI, Qwen Code, and Kimi CLI |
| Browser Automation | Chromium, Playwright, and full desktop (VNC) sandboxes for GUI agents |
| Code Execution | Code Interpreter SDK for multi-language code execution with structured result output |
| Agent Evaluation | Reproducible, ephemeral sandboxes for scoring and benchmarking agent outputs |
| RL Training | Containerised training workloads (e.g. DQN CartPole) with checkpoint and summary output |
| Remote Development | code-server (VS Code Web) and desktop environments running inside a sandbox |
Project Structure
The repository is organised into focused directories. Each directory ships as an independent component that can be consumed separately.| Directory | Description |
|---|---|
sdks/ | Multi-language SDKs — Python, Java/Kotlin, TypeScript/JavaScript, C#/.NET, Go |
specs/ | OpenAPI specs and lifecycle specifications |
server/ | Python FastAPI sandbox lifecycle server |
cli/ | osb command-line interface |
kubernetes/ | Kubernetes deployment manifests and examples |
components/ | Execution daemon (execd), ingress proxy, and egress control sidecar |
sandboxes/ | Runtime sandbox image implementations |
examples/ | Runnable example code for agents, browsers, ML, and more |
Where to Go Next
Quickstart
Start the server, install the Python SDK, and run your first sandbox in under 10 minutes.
SDKs Overview
Explore the full SDK surface across all supported languages.
API Reference
Browse the OpenAPI-generated reference for every lifecycle and execution endpoint.
CLI Reference
Full command reference for the
osb CLI tool.