Run and supervise agents
See AI support agents for setup and Run the full support loop for the journey from ticket intake to verified resolution.
The desktop for investigation and intervention
When a case needs an operator, open the relevant evidence and tools in the desktop. Navigation and channels live on the left. The active conversation stays central. Terminals, files, dashboards, procedures, settings, and other artifacts open in a tabbed workbench beside it. Live tasks and account state remain visible around the edges. The desktop’s local tools and credentials are separate from the cloud agent’s execution environment.The left sidebar
The fixed navigation at the top of the sidebar contains:
Below the fixed navigation, Studio shows recent dashboards, the channel switcher, and the conversation list. The lower sidebar surfaces running activity and your organization/account controls.
The Inbox is separate from the conversation chat board. Open a waiting task to review the worker’s evidence and respond or hand it off. The browser-based Work area also provides My approvals, Boards, Outcomes, and Notifications; see Digital workers.
Channels and conversations
The channel switcher controls which conversations, grounding, members, and tool policy apply to new work. The organization switcher controls the larger trust boundary. They are different scopes. The chat list supports pinning, renaming, archiving, and work status. Choose View as board from the channel menu to group conversations into Processing, In progress, Cancelled, and Done. These conversation statuses do not control worker runs or external tickets. Check the selected channel in the switcher; shared conversations also show a channel chip in their header. The composer gives you:- A plus menu for attachments and context.
@mentions for hosts, files, procedures, connectors, MCP servers, and tools.- Default, Ask, and Planning modes.
- The optional Autopilot permission.
- Model selection for the conversation, where organization policy allows it.
- Queue or steer behavior while Copilot is already running.
Conversation and artifacts workspace
The conversation is the operating thread. Tool calls, approvals, plans, sub-agent progress, generated UI, cost details, and completion state all stay in its transcript. Studio opens the things a conversation works with in the artifacts workspace: terminals, host editors, browser sessions, remote desktops, files, reports, code, dashboards, generated apps, procedures, settings, calls, and usage. Toggle the artifacts workspace withMod+Shift+B. Toggle the bottom panel with Mod+J.
Tabs and editor groups
Tabs can open as previews, stay open, be pinned, or move between groups. Split the workbench right or down when two surfaces need to remain visible.
Lock a group when a live terminal, call, or dashboard must not be replaced by the next preview. Pin an editor when it should survive routine navigation.
Search and commands
PressMod+K to search Studio entities and registered commands. Results can include hosts, conversations, procedures, files, dashboards, settings, and actions. Search the noun when you know the object and the verb when you know the action.
Examples:
edge-03to find a host or conversation mentioning it.billingto open Billing or Usage.split editorto find a workbench action.key chainto open credential management.
Activity and status
The activity stack shows work that continues outside the currently visible surface: agentic chat runs, sub-agents, browser sessions, calls, procedure runs, scheduled wakeups, sync, and other background work. Open an item to return to its owning conversation or workspace. The status bar exposes current health and account state, including running activity, connectivity, sync, usage balance, Studio Remote, and context-specific session indicators. When a task appears stuck, open its activity detail before retrying.Useful global shortcuts
See Keyboard shortcuts for the full workbench list and custom procedure triggers.
Handle a case that needs an operator
Use a human task to bridge a cloud workflow and work that requires an operator’s desktop access. Return the findings through the task so the agent can continue with evidence. The handoff does not transfer desktop credentials or remote-session access to the agent. Required human tasks must resolve before the run is queued to resume.- Confirm the organization and channel.
- Open or mention the exact host, file, procedure, connector, or dashboard.
- Start in Ask or Planning when the target is unfamiliar.
- Gather evidence in the appropriate structured surface.
- Review approvals in the conversation before state changes.
- Save the result as an artifact, dashboard, private memory, reviewed shared fact, or procedure when it should outlive the chat.
- Return the verified findings and action taken to the worker’s human task or the support system. Update the customer and case status through the configured workflow; moving a chat-board card does not close a ticket.
Related
Channels and the chat board
Define context and policy, then track conversations by status.
AI Copilot
Choose a mode, attach context, review tools, and steer a run.