openkruise/agents

What it solves

OpenKruise Agents simplifies the infrastructure challenges of running AI agents within Kubernetes. It provides a standardized, vendor-neutral way to manage the isolated environments, persistent workspaces, and resource-intensive training jobs that AI agents require, bridging the gap between AI development and cloud-native infrastructure.

How it works

It functions as a management layer for AI agent sandboxes on Kubernetes. It enables rapid resource provisioning and dynamic resizing to keep costs low. The system supports advanced features like sandbox hibernation and checkpointing—which can save the state of memory, storage, and even GPU memory—and handles user session management and traffic routing to reduce reliance on standard Kubernetes services. Developers can interact with these environments using either Kubernetes CRD APIs or the E2B SDK.

Who it’s for

This project is designed for infrastructure engineers managing AI workloads and AI algorithm scientists who need reliable, scalable, and isolated environments to run agentic tools, research notebooks, or training jobs.

Highlights

• Provides isolated, secure sandboxes for AI agent tool usage and development.
• Supports stateful hibernation, allowing sandboxes to save and resume memory and GPU states.
• Offers flexible API support, including compatibility with E2B SDKs and Kubernetes-native APIs.
• Enables efficient resource pooling and dynamic scaling to optimize infrastructure costs.
• Facilitates complex workloads like reinforcement learning and big data training with rapid startup times.