Agent Sandbox

Agents in Sandbox — OpenClaw

Run always-on agent environments with OpenClaw inside Agent Sandbox for persistent, long-running workloads.

Overview

Some agent workloads are not short-lived tasks but persistent services that run continuously. OpenClaw (formerly Moltbot) is an always-on agent that runs inside Agent Sandbox, benefiting from the sandbox’s stable identity, persistent storage, and Kubernetes-native lifecycle management.

This is the always-lived pattern: the sandbox runs indefinitely as a persistent service, responding to requests and maintaining state across restarts.

Why Use a Sandbox for Always-On Agents?

  • Stable identity — Each sandbox has a stable hostname and network identity, so the agent is always reachable at the same address.
  • Persistent storage — Sandboxes can mount persistent volumes so the agent’s data survives pod restarts.
  • Web UI and CLI access — Agents like OpenClaw expose a web interface and support CLI operations, accessible via port-forwarding.
  • Token-based authentication — Secure access to agent interfaces through gateway authentication.
  • Lifecycle management — The agent-sandbox controller handles pod creation, restarts, and scheduled deletion without manual intervention.

Getting Started

See the OpenClaw Sandbox example for a complete walkthrough covering image loading, token generation, sandbox deployment, web UI access, and CLI operations.

Last modified April 23, 2026: Docs feature use cases (#652) (0840ee5)