Agent Pods are persistent cloud workspaces where your AI agents run — around the clock, together, and under the controls your business sets. Today they are live in every SkyDeck.ai workspace.
The way teams use AI agents has outgrown the chat window. The strongest coding agents — Claude Code, OpenClaw — do their most valuable work over long stretches: a big refactor, a full test run, a migration that takes hours. A browser tab cannot hold that. Close the laptop and the work is gone. Agent Pods close the gap by giving an agent a home that keeps running. It is the workflow your engineers already love, made shared, secure, and manageable by the business.
A pod is a persistent cloud workspace you launch from GenStudio. Pick your agent — Claude Code, OpenClaw, or a custom image — and start work in a real terminal. The main terminal survives your disconnects: close the tab, shut the laptop, lose the wifi, and the session keeps going. A built-in browser pane sits beside the terminal so you, or the agent, can drive a real browser and preview web apps on the pod's exposed ports. When a pod goes idle it auto-sleeps to save spend, then resumes exactly where it left off.
Nothing about a pod is disposable. Everything in /workspace persists across sessions, and Claude Code pods also keep ~/.claude, so your repository, your dependencies, your credentials, and your history are all there when you come back. Wake a pod a week later and the agent picks up where it stopped — no re-cloning, no re-auth, no lost context.
Teams reach for pods in three shapes — and the same pod often plays all three roles over its life.
Run around the clock. Kick off a long Claude Code session or a detached OpenClaw gateway and let it work while you sleep. The main terminal outlives your connection, auto-sleep keeps an idle pod from burning money, and a budget cap means it can never run past what you have allowed. Come back in the morning to finished work instead of a prompt waiting for your next message.
Operate together, one Driver at a time. A pod is not a private tab. Invite teammates in, and exactly one Driver holds the keyboard while everyone else observes live and read-only. Request or give control in a click and the keyboard passes cleanly — to a reviewer, a new hire, or the next time zone. Real code review, onboarding, and follow-the-sun operation happen inside the same terminal, without the collisions of a shared login.
Deploy to the whole team. Ship a standard system pod to every member, or build a custom image and publish it to the groups that need it. Scope which groups can launch which pods and agents, and keep the entire fleet under one set of guardrails. Everyone starts from the same ready-to-run setup instead of a page of instructions and a lost afternoon.
Always-on agents are only safe if someone can bound them, so governance is not an add-on here — it is where pods start. The Control Center runs the whole fleet. Set a monthly budget and choose to block, warn, or allow overage when it is reached. Cap spend per member and per group. Limit how many pods run at once to keep load and cost predictable. Let idle pods auto-sleep. And watch every running pod — its driver, its agent, its current spend — from one screen.
Access stays governed too. Encryption in motion and at rest, SSO, DLP, and RBAC decide who can reach which models and agents. And because SkyDeck is a front end for your whole AI stack, pods carry no vendor lock-in: reach OpenAI, Anthropic, Google, and dozens more models through one interface, and move budget and access as your priorities change.
Agent Pods are available now in every SkyDeck.ai workspace, governed by the Control Center from the very first run. Launch one in a free workspace, drop in Claude Code or OpenClaw, and put an agent to work tonight. Read the Agent Pods guide to go deeper, or explore the full picture on the Agent Pods page.