Skip to the collection

agent-native · Code Execution · Verified 2026-10-13

Agent Sandbox (Kubernetes SIG)

Listed on The Agent-Native Index

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

What it does

Kubernetes SIG Agent Sandbox (SIG Apps) is a declarative API for isolated, stateful, singleton workloads — the kind of long-running, identity-stable container AI agent runtimes and RL evaluators need. The core Sandbox CRD gives a stable hostname, persistent storage, and lifecycle (create, scheduled delete, pause, resume). Extensions add reusable templates, warm pools, and claim-based allocation.

Official scope: it orchestrates sandboxes; it does not implement gVisor or Kata itself. Pods select those isolation backends through RuntimeClass. Homepage use cases include short-lived code execution, coding agents, computer-use desktops, CI, and always-on OpenClaw environments.

Distinctness: this entry is the in-cluster CRD/controller. The catalog already lists hosted Agent Sandbox (agentsandbox.co) — a different product with URL onboarding. OpenSandbox is a separate sandbox runtime that can sit on Kubernetes and may consume this CRD; it is not this SIG project.


Agent-facing interfaces

How agents onboard

Interaction pattern: SDK + Kubernetes CRDs

pip install k8s-agent-sandbox

See the full onboarding section in the dossier.

Protocol surface

Interface Detail
Kubernetes API Sandbox, SandboxTemplate, SandboxClaim, SandboxWarmPool
Python SDK pip install k8s-agent-sandbox — PyPI
Go SDK go get sigs.k8s.io/agent-sandbox/clients/go/sandbox@latest
Docs https://agent-sandbox.sigs.k8s.io/docs/

Verification

Last verified 2026-10-13.

Sources: