> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agentruntime.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Command Center

> Review approvals, monitor active runs, and recover from failures in the AgentRuntime run inbox.

**Command Center** is your workspace-wide run inbox. Open it from the Console sidebar at **Command Center** (`/command-center`).

Use it when you need to approve human tasks, pause or stop runs, or triage recent failures — without opening every workflow in Workflow Studio.

## What you see

Command Center organizes runs into four sections:

| Section                 | What it shows                                                         |
| ----------------------- | --------------------------------------------------------------------- |
| **Awaiting approval**   | Runs paused on `human_task` steps with pending approve/reject actions |
| **Stopping gracefully** | Runs in `draining` status — graceful stop in progress                 |
| **Running now**         | Live, queued, paused, and starting runs (excluding draining)          |
| **Failed recently**     | Runs that failed or stopped in the last 24 hours                      |

The page header shows badges when action is needed:

* **N pending** — human tasks waiting for approval
* **N draining** — graceful stops still winding down

Data refreshes automatically every 3 seconds when there are pending approvals or active runs, otherwise every 30 seconds.

## Awaiting approval

When a workflow hits a `human_task` step, the run pauses and tasks appear here grouped by workflow and run.

For each pending task you see:

* Workflow name and run ID (first 8 characters)
* Task type (for example, `approval`)
* Step ID that triggered the pause

### Approve or reject

<Steps>
  <Step title="Review the task">
    Read the task type and step ID. For full payload context (draft email, proposed data changes), click **Review in studio** to open Workflow Studio on the run context tab.
  </Step>

  <Step title="Decide">
    Click **Approve** to continue the run with an approval result, or **Reject** to stop the branch with a rejection result.
  </Step>

  <Step title="Confirm the run resumes">
    After approval, the task disappears from Command Center and the run continues. Check **Running now** or open the run in Studio to watch downstream steps.
  </Step>
</Steps>

<Note>
  Approve/reject here is the same action as completing a task in Workflow Studio or via `POST /v1/workflows/{id}/runs/{runID}/tasks/{task_id}/complete`. See [Human tasks](/workflows/human-tasks).
</Note>

## Running now

Active runs appear with status badges:

| Status                                  | Meaning                                                       |
| --------------------------------------- | ------------------------------------------------------------- |
| **running**                             | Steps are executing                                           |
| **paused**                              | Operator paused the run — no new steps schedule until resumed |
| **pending** / **queued** / **starting** | Run is waiting to start or between states                     |
| **draining**                            | Graceful stop — shown in the separate draining section        |

### Run controls

From **Running now** (and **Stopping gracefully**), you can:

| Action               | Effect                                                           |
| -------------------- | ---------------------------------------------------------------- |
| **Stop (immediate)** | Halt now — cancels in-flight MCP calls; default stop mode        |
| **Stop (graceful)**  | Enter `draining` — let active steps finish, block new scheduling |
| **Pause**            | Pause scheduling; in-flight steps may complete                   |
| **Resume**           | Continue a paused run                                            |

Click **Open run** to jump to Workflow Studio's event log for that run.

<Warning>
  Immediate stop can leave downstream steps unexecuted. Graceful stop waits for in-flight work but never starts steps that were only pending. Both are terminal — you cannot resume a stopped run.
</Warning>

## Stopping gracefully

When you choose graceful stop, the run moves to **Stopping gracefully** until all in-flight steps complete. The badge reads **Draining**.

If a drain takes too long, escalate to **immediate stop** from the same row. A drain timeout surfaces as `DRAIN_TIMEOUT` in the failure log.

## Failed recently

Runs that **failed** or were **stopped** in the last 24 hours appear here. Open the run in Workflow Studio to read step-level error messages and `failure_code` values.

Common failure codes:

| Code                       | Meaning                                         |
| -------------------------- | ----------------------------------------------- |
| `MCP_TOOL_FAILED`          | MCP tool returned an error                      |
| `LLM_REQUEST_FAILED`       | LLM call failed (key, model, or provider error) |
| `INSUFFICIENT_CREDITS`     | Workspace out of credits                        |
| `RUN_STALLED_OR_TIMED_OUT` | No progress within the timeout window           |
| `RUN_STOPPED`              | Operator stopped the run                        |

See [Troubleshooting](/platform/troubleshooting) for recovery steps.

## When a run looks stuck

Work through this checklist:

1. **Check status in Command Center** — Is it `paused`, `draining`, or waiting on approval? Complete the human task or resume if paused.
2. **Open the run in Studio** — Read the event log for the last step started. A step stuck in `running` may be waiting on a slow MCP call or LLM response.
3. **Try pause then stop** — Pause prevents new steps. If still stuck, use immediate stop.
4. **Check connections** — MCP binding failures often show as `MCP_TOOL_FAILED` on the first tool step. See [Troubleshooting](/platform/troubleshooting).
5. **Check credits** — `INSUFFICIENT_CREDITS` blocks new runs and can fail in-flight billing gates.

Runs waiting on human approval are **not stuck** — they are designed to pause until you approve or reject.

## API equivalents

| Console action        | API                                                                                                |
| --------------------- | -------------------------------------------------------------------------------------------------- |
| List pending tasks    | `GET /v1/tasks/pending`                                                                            |
| Complete task         | `POST /v1/workflows/{id}/runs/{runID}/tasks/{task_id}/complete` with `{ "approved": true\|false }` |
| List tenant runs      | `GET /v1/runs`                                                                                     |
| Stop / pause / resume | `POST /v1/workflows/{id}/command` with `stop`, `pause`, or `resume`                                |

## Related docs

* [Runs and observability](/workflows/runs-and-command-center) — run lifecycle, failure badges
* [Run recovery (roadmap)](/workflows/run-recovery) — planned continue-from-failure
* [Human tasks](/workflows/human-tasks) — authoring approval gates
* [Workflow patterns](/workflows/patterns) — approve-then-send and error handling