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

# Agent Run

> Execute and monitor automations in real-time

## The Execution Interface

When you run an automation, you're taken to the Agent Run page. This is where you watch, interact, and control the execution.

## Interface Layout

<CardGroup cols={2}>
  <Card title="Video Panel" icon="video">
    Live stream of the Windows desktop
  </Card>

  <Card title="Chat Panel" icon="comments">
    Action-by-action log and HITL controls
  </Card>

  <Card title="Progress Bar" icon="bars-progress">
    Current stage of execution
  </Card>

  <Card title="Control Bar" icon="sliders">
    Play, pause, cancel buttons
  </Card>
</CardGroup>

## Live Video Streaming

The left side shows a real-time video of the Windows desktop:

* **WebRTC** powered - Low latency, high quality
* **Full screen** option available
* **Screenshot** button to capture current view

<Info>
  Video streaming requires WebRTC support in your browser. Most modern browsers work fine.
</Info>

## Chat Panel

The right side shows the conversation with the AI agent:

```
Agent: Starting automation "Weekly Sales Report"
       Connecting to driver machine...

Agent: Connected. Opening Excel application.
       [Screenshot captured]

Agent: I'm about to click "File > Open"
       [Approve] [Modify] [Skip] [Cancel]

You: [Clicked Approve]

Agent: Opening file dialog. Navigating to Sales folder...
```

## HITL Controls

When the agent pauses for approval, you'll see these options:

<Tabs>
  <Tab title="Approve">
    **When to use:** The action looks correct

    Click **Approve** to let the agent proceed with the planned action.
  </Tab>

  <Tab title="Modify">
    **When to use:** The action is close but needs adjustment

    Enter a modification like:

    * "Click the blue button instead"
    * "Type 'Q1 2024' not 'Q1 2023'"
    * "Wait 5 seconds first"
  </Tab>

  <Tab title="Skip">
    **When to use:** This step isn't needed

    The agent will skip the current action and move to the next one.
  </Tab>

  <Tab title="Cancel">
    **When to use:** Something is wrong, stop everything

    Immediately stops the execution. No further actions will run.
  </Tab>
</Tabs>

## Progress Stages

The progress bar shows where you are in the execution:

| Stage          | Description                       |
| -------------- | --------------------------------- |
| **Queued**     | Waiting for an available driver   |
| **Connecting** | Establishing connection to driver |
| **Running**    | Actively executing steps          |
| **Waiting**    | Paused for HITL approval          |
| **Completed**  | Finished successfully             |
| **Failed**     | Encountered an error              |
| **Cancelled**  | Stopped by user                   |

## Screenshots

Throughout execution, Granite captures screenshots:

* **Before** each action
* **After** each action
* **On errors** for debugging
* **Final state** when complete

View screenshots in the slideshow below the video panel.

## Recordings

Every execution is recorded:

* Full video from start to finish
* Downloadable after completion
* Useful for auditing and review

## Execution Controls

<CardGroup cols={3}>
  <Card title="Pause" icon="pause">
    Temporarily stop execution (resume later)
  </Card>

  <Card title="Cancel" icon="stop">
    Stop execution permanently
  </Card>

  <Card title="Full Screen" icon="expand">
    Maximize the video panel
  </Card>
</CardGroup>

## Handling Errors

If something goes wrong:

1. **Video shows the error state** - See exactly what happened
2. **Chat explains the error** - Agent describes what went wrong
3. **Options to recover**:
   * Modify the approach and retry
   * Skip the problematic step
   * Cancel and investigate

<Warning>
  If the agent repeatedly fails on the same action, it may need a different approach. Try modifying the instruction or creating an RPA script for more control.
</Warning>

## After Execution

When the run completes:

* **Summary** - Success/failure, duration, action count
* **Screenshots** - Browse all captured images
* **Recording** - Download the video
* **Logs** - Detailed trace of every step

## Tips for Effective Monitoring

<AccordionGroup>
  <Accordion title="Keep the video visible">
    The video shows exactly what the agent sees. If something looks wrong, you can catch it immediately.
  </Accordion>

  <Accordion title="Read the chat carefully">
    The agent explains what it's about to do before doing it. This is your chance to catch mistakes.
  </Accordion>

  <Accordion title="Use Modify, not Cancel">
    If an action is close but not quite right, modify it instead of cancelling. You'll preserve progress.
  </Accordion>

  <Accordion title="Review recordings later">
    Even if execution succeeds, watch the recording to understand how the automation works. You might find optimization opportunities.
  </Accordion>
</AccordionGroup>

## Keyboard Shortcuts

| Shortcut | Action                 |
| -------- | ---------------------- |
| `Enter`  | Approve current action |
| `Escape` | Cancel execution       |
| `Space`  | Pause/Resume           |

## Next Steps

<CardGroup cols={2}>
  <Card title="View Run Logs" icon="list" href="/dashboard/run-logs">
    See execution history
  </Card>

  <Card title="HITL Deep Dive" icon="hand" href="/agent-execution/hitl-workflows">
    Master human oversight
  </Card>
</CardGroup>