OpenSandbox

OpenSandbox gives developers isolated environments for AI agents that need to execute code or interact with browsers and developer tools. It runs on your own infrastructure, locally with Docker or on a Kubernetes cluster.

• Agent workspaces: Agents can run shell commands, read and write files, and use code interpreters while applications receive streamed outputs, logs and metrics.
• Coding and browser workloads: Sandboxes can run Claude Code, Gemini CLI and Codex, host Chrome and Playwright automation, or provide VS Code Web and desktop environments.
• Network and credentials: Per-sandbox policies control outbound access, a gateway routes inbound traffic, and a credential vault lets agents call external services without exposing real credentials to sandbox workloads.
• Sandbox lifecycle: Applications can create, monitor, pause, resume and terminate environments; Firecracker sandboxes retain memory and disk state during pauses.
• Agent evaluation: Resource pools and batch creation support agent evaluation and reinforcement learning workloads.

A shared API covers local and cluster deployments. SDKs support Python, Java/Kotlin, JavaScript/TypeScript, C#/.NET and Go. The MCP server gives clients such as Claude Code and Cursor access to sandbox creation, command execution and text file operations.

OpenSandbox is open source under Apache 2.0. Local execution requires Docker, and the local runtime and examples require Python 3.10+.