OpenSandbox
Isolated sandbox runtime for AI agents on Docker or Kubernetes
What it is
OpenSandbox runs AI agent workloads in isolated environments where they can execute code, run shell commands, manage files, and drive browsers or desktops. It starts locally on Docker and scales to Kubernetes, with Firecracker microVMs that support pause and resume, per-sandbox egress policies, and a credential vault that keeps real secrets out of sandboxes. Teams building agent products, evaluation harnesses, or RL training pipelines reach it through SDKs in Python, TypeScript, Java, C#, and Go, a CLI, or MCP.