Agent Sandbox provides a secure, and isolated execution layer to safely deploy autonomous AI agents on Kubernetes that generate and run untrusted code at scale.
Users can leverage Agent Sandbox to safely execute arbitrary, untrusted code within an isolated environment, enabling applications such as stateful code interpretation, agentic web browsing, computer use, and sophisticated data analysis.
The Agent Sandbox is specifically designed for interoperability by defining a standardized Kubernetes API that decouples the execution layer from the underlying technology. This abstraction allows the sandbox to support various backends, such as gVisor, and Kata Containers, enabling users to choose the isolation technology that best fits their security, performance, or workload requirements.
Contributions welcome!
We do a Pull Request contributions workflow on GitHub. New users are always welcome!