Skip to main content
When a request is large enough, the agent leading your chat splits it into bounded tasks and hands each one to a temporary subagent — a short-lived helper agent that does one job and finishes. The chat side panel is where you watch that team work, answer anything that needs you, and replay every command the agents ran.

​
Why the side panel

  • Live visibility. Each helper agent has its own transcript you can open while it works — you see the team, not just the final answer.
  • Nothing waits silently. An agent that needs your approval or review raises a visible count instead of stalling the conversation unnoticed.
  • Recoverable work. Failed work gets a one-click Retry; a running agent you no longer want can be stopped without losing what it already finished.
  • A full command record. The Console tab keeps every command run in the conversation, step by step, so you can scroll back through exactly what happened.
Open the panel any time with the Show panel button at the top right of a chat. On a phone, use the Panel tab at the bottom of the screen. The panel opens on its own only when a reply you are watching starts new agent work — reopening a conversation later never pops it open.

​
The Agents tab

The Agents tab appears whenever the conversation has an agent team. It lists every helper agent with its name, current status, and a one-line summary. Select a row to open that agent’s work: its assigned objective and a live feed of what it is doing. A newly started agent shows a loading state until its first activity arrives.

​
When an agent needs you

Two statuses ask for your attention, and they ask different questions:
  • Needs approval — the agent wants permission before running an action. Open the agent’s work and answer the Approve the requested action? card, which states the action and the reason. Show in chat jumps to the same request in the transcript.
  • Needs review — an action already ran (or may have), and the agent wants you to confirm the outcome before it continues.
While agents are waiting, a count badge rides the Show panel button and the Agents tab — on a phone, the Panel tab carries it. CloudThinker never switches your view for you: the badge tells you someone is waiting, and you choose when to look. Which actions pause for approval in the first place is governed by your workspace’s autonomy mode and Tool Permissions.

​
Retry and stop

Retry appears only on an agent whose status is Failed. It sends a fresh request to the agent leading the conversation, restating the failed objective — the failed helper is not resumed mid-task, so the lead agent decides how to approach it again. Retry is unavailable while a reply is still streaming, and stopped work has no Retry. To stop a running agent, open the menu on its row and select Stop agent. Stopping affects only that one agent: work it already finished is kept, and the rest of the team keeps going.

​
The Console tab

The Console tab is available in every conversation and holds the complete command history — each command the agents ran, with its output, recorded as numbered steps.
  • The step counter shows where you are, such as 12/40.
  • Use the Previous step and Next step arrows, or drag the progress bar, to move through history.
  • Jump to Live returns you to the newest activity while the conversation is still running.
The same panel also collects the conversation’s outputs — reports, dashboards, generated images, artifacts, and files — each on its own tab that appears once there is something to show.

​
FAQ

A reply you were watching started new agent work, so the panel opened to show the team. That is the only time it opens on its own — reloading or revisiting a conversation never triggers it.
No. The waiting agent parks on its request while the rest of the team continues. The count badge on the panel button tells you how many agents are waiting.
Retry exists for work that failed on its own. Stopping is your decision, so CloudThinker does not offer to redo it — send a new message if you want the task attempted again.
No. A newer run took over the same task, which is how continued work is handed forward. The replaced agent’s transcript stays readable.

Agents

The five specialist agents and how they coordinate

Auto Mode

How Manual and Auto decide which agent actions pause for you

Chat features

Attachments, message queueing, and other conversation mechanics

Capabilities

What agents produce — dashboards, reports, alerts, and images