agent-sandbox

Why we included this project

Production AI agents usually live a long time. They hold working memory, keep a stable hostname so other services can reach them, and carry state that has to survive a restart. Agent Sandbox makes that kind of workload a native Kubernetes resource instead of a workaround: its Sandbox custom resource gives a single pod a stable identity, persistent storage, and a controller that handles creation, hibernation, resuming, and scheduled deletion through declarative YAML. Isolation itself is handed off to secure runtimes like gVisor or Kata Containers through RuntimeClass, which is the part that matters when you are running untrusted, LLM-generated code in a shared cluster. For teams operating agent infrastructure or isolated dev environments, it beats wiring up StatefulSets, Services, and volumes by hand.