Skip to content

Run AI Agents Around the Clock

An AI agent is most useful when it can keep working after you stop watching it. On your laptop that's fragile: the lid closes, the Wi-Fi drops, the machine reboots — and the agent dies with it. A pod breaks that dependency: it's a persistent Linux machine in the SkyDeck cloud that keeps its state, runs independently of your device, and is reachable from any browser.

Long-running coding sessions with Claude Code

The Claude Code pod comes with the agent, git, the GitHub CLI and a durable /workspace volume already set up.

  1. Launch the Claude Code pod from the pod catalog. You land in the pod's main terminal — the session that persists.
  2. Clone or open your repository under /workspace (your admin may have configured the pod to clone it for you), then run claude and put the agent to work — a big refactor, a test-suite cleanup, a migration.
  3. Walk away. Close the tab, shut your laptop, commute. The main terminal — and the agent working in it — lives in the pod, not on your device, and keeps running after you disconnect. The only clock that matters is the pod's idle auto-sleep (below).
  4. Reattach from anywhere. Open the pod on any machine and the same session — scrollback, running agent and all — is waiting.

Because the Claude Code pod persists both /workspace and the agent's own memory and login (~/.claude) across sleep and resume, a project can span days or weeks of sessions: each time you resume, the agent still has its context, settings, and skills. Open extra terminal tabs (up to 10) to run several agent tasks side by side.

Always-on general agents (OpenClaw and friends)

Coding is one shape of agentic work. General agents — like OpenClaw, which chats with you over your messaging apps and acts on your behalf — are designed to run continuously as a service. That's exactly what a pod is for:

  1. Ask your admin to add a custom pod with the OpenClaw agent preinstalled (it's one of the built-in agent choices when creating a custom pod), or start from any pod and install your own agent in /workspace.
  2. Start the agent as a background service, so it doesn't depend on a terminal tab staying open:
tmux new -d -s openclaw 'openclaw gateway'

A detached process keeps running as long as the pod is awake, independent of any terminal — the right shape for a service. (Work in the pod's main terminal also survives your disconnects, but keep that for interactive sessions; extra terminal tabs are tidied up once everyone's left them.) 3. Check on it any time from the messaging channel the agent connects to, or open a terminal in the pod: openclaw gateway status, or tmux attach -t openclaw to see it live.

Keeping it awake (and affordable)

Two organization-level settings decide how "24/7" your pod really is — both live in Control Center → Pods (spend controls):

  • Idle auto-sleep puts a pod with no terminal activity to sleep (default 20 minutes). For always-on agent workloads, ask your admin to raise it — or turn it off so pods run until slept manually.
  • Budgets and caps are the counterweight: a monthly organization budget, per-member caps, and a concurrent-pod limit keep an always-on fleet from becoming an open-ended bill. A pod that's asleep costs only storage.

Info

A practical pattern: let ordinary dev pods keep the default auto-sleep, run always-on agents in their own pods, and size the per-member caps so an around-the-clock pod fits inside them. The Control Center shows each pod's hourly price before you launch it.