Skip to main content
The Agency Command Center is the central operations cockpit for the Agent Lab platform. From a single page you can launch DAG (directed acyclic graph) workflow runs into the execution engine and watch the status of every swarm agent. You can also approve or reject runs that pause for human sign-off and hand tasks to the Ops Orchestrator in plain English. Use the Command Center when you are the operator on duty. It replaces jumping between the workflows list, the runs log, and the agents roster with one live view of the whole operating system.

​
Prerequisites

  • An Agent Lab account. The Command Center requires authentication. If you open it while signed out, Agent Lab redirects you to the login page.
  • At least one deployed workflow if you want to dispatch runs. Without workflows the launcher shows “No workflows configured.”

​
Open the Command Center

Sign in, then navigate to /command-center in the web app. The header confirms the page is connected to the live runtime and shows the AI Studio mobile bridge status. Four header actions are available:
  • Open Ops Agent jumps to the Ops Agent page at /ops-agent for cleanup and triage chat sessions.
  • Refresh re-fetches workflows, agents, and runs on demand.
  • 1-Click Sync All calls POST /api/sync/all to align the desktop HTML, repo Markdown, and OS state in one action.
  • Run data also refreshes automatically every 5 seconds, so the approval queue stays current without manual refreshes.
Below the header, the Operating Focus & Today’s Top Actions card lists the current operational directives derived from the URC Agency Operating Manual. Every session starts from the same priorities.

​
The four panels

​
Deployable workflows & SOP playbooks

The playbook launcher lists every active DAG workflow with its name, description, trigger type, step count, and success rate. Click Execute Run on any card to dispatch that workflow into the execution engine immediately. Each workflow card expands. Click the chevron to reveal the DAG execution node sequence: every step’s title, type, action prompt, and assigned swarm agent. Use the expanded view to inspect exactly what a playbook will do before you dispatch it. Each dispatch calls the run trigger endpoint with the Command Center recorded as the trigger source:
A toast confirms when the run is initiated, and the new run appears in the runs feed on the next 5-second refresh.

​
Adjust workflow steps

Click Adjust Steps on any workflow card to open the step editor. The editor lets you fine-tune the DAG execution graph without recreating the workflow:
  • Edit the workflow’s name, description, and trigger type. Trigger types are manual execution, CRON schedule, webhook or inbound event, and event stream.
  • Add, remove, and reorder step nodes.
  • Set each step’s type: trigger, agent, guardrail, or destination node.
  • Rewrite each step’s action prompt and reassign its swarm agent.
Click Save Workflow Adjustments to apply your changes. The Command Center updates the workflow metadata with PATCH /api/workflows/{workflowId} and replaces the step graph with a single request:
If a workflow has no step nodes configured yet, the editor pre-fills a three-node template (trigger, agent, and destination) so you can start from a working graph.

​
Swarm agents

The swarm matrix shows every agent node registered on the platform. Each row displays the agent’s name, role, base model, and uptime. A pulsing green dot marks agents with active status; a gray dot marks idle or offline nodes. Use this panel to confirm the swarm is healthy before dispatching high-volume work.

​
Approval queue

Some workflow runs pause mid-execution for human-in-the-loop sign-off and enter the paused_for_approval state. The approval queue surfaces every paused run with its workflow ID and run ID, plus two actions:
  • Approve resumes the run from the paused step (POST /api/runs/{runId}/approve).
  • Reject cancels the run (POST /api/runs/{runId}/reject).
When nothing is pending, the panel shows “All systems autonomous.” Check this queue first each session. Paused runs block their downstream steps until you act.

​
Ops Orchestrator terminal

The terminal at the bottom of the page is a natural-language command line into the Orchestrator, powered by Google Gemini 2.5 Flash. Type an operational task and click Send. For example:
The Orchestrator can dispatch cross-department tasks, synthesize new DAG workflows, or answer questions about real-time platform state. Each message posts to POST /api/orchestrator/chat and the reply streams back into the message feed with a timestamp. If the Orchestrator cannot complete a command, it reports the error in the feed and continues operating in fallback mode. When the Orchestrator synthesizes a new workflow, its reply includes a proposal card with the proposed name, department code, and description. Click Approve & Deploy DAG to send the proposal to POST /api/workflows/deploy. The deployed workflow then appears in the playbook launcher, where you can adjust its steps or execute a run.

​
When to use each panel