Skip to content

Deploy Agents to Your Whole Team

The hard part of adopting AI agents isn't running one — it's running them for everyone: the same environment, the right access, and a bill you can predict. With pods, an organization admin builds the agent environment once and every member launches their own copy of it — no per-laptop setup, no configuration drift, no credentials scattered across machines.

All of this happens in Control Center → Pods (admins and owners; sign in at admin.skydeck.ai).

1. Start with the built-in agent pod

The fastest rollout is the system Claude Code pod — a ready-made coding-agent environment (Ubuntu, git, GitHub CLI, Claude Code, durable /workspace). Toggle it on and choose who sees it: everyone, or specific groups. Members find it in their pod catalog the next time they open the workspace.

2. Make it your environment with a custom pod

When your team needs more than the base — your repositories, your toolchain, a different agent — create a custom pod:

  • start from a SkyDeck base image, pick the agent to preinstall (Claude Code, OpenClaw, Hermes, JupyterLab, or none), point it at a repository to clone into /workspace, add a startup command and exposed ports, and pick a size (Small to X-Large) — or bring your own image for a fully baked environment;
  • scope it to the groups that should have it — engineering gets the coding pod, analysts get the JupyterLab pod, support gets the OpenClaw pod;
  • update it centrally: push a new image tag and edit the pod — members get the new environment on their next launch.

Everyone who launches the pod gets an identical, isolated machine — same tools, same agent, their own workspace.

3. Put guardrails around the spend

Agent fleets fail rollouts when the bill surprises someone. Set the policy up front in Control Center → Pods:

  • Monthly budget for the whole organization, and what happens on overage — block new launches, warn, or allow and track.
  • Per-member cap by default, with per-group caps where one team needs more.
  • Concurrent pods per member, so nobody fans out a surprise fleet.
  • Idle auto-sleep (default 20 minutes) so forgotten pods stop billing on their own — raise or disable it for teams running always-on agents.

Each pod shows its hourly price (by size, plus a managed-agent surcharge) before launch; running pods bill compute per hour, sleeping pods only storage.

4. Watch it run

The same Control Center page is your fleet view: current month spend against budget, who's spending what, and every running pod in the organization — which an admin can sleep, stop, or delete directly if something's been left on or gone wrong.

Info

A rollout that works: enable Claude Code for one pilot group with the default caps → add a custom pod with your repositories and the agents each team needs → widen availability group by group, raising caps where the usage justifies it. Because pods are centrally defined, "upgrade the team's agent" is an image-tag edit, not an IT project.