# Start an automation

Start a workflow agent's run by hand from its page or from a conversation, give it inputs or files, follow its steps, and stop or approve it while it runs.

An automation is a run of a workflow agent: instead of answering in a chat, the agent works through its steps and leaves what it produces in Artifacts. Most runs start on a schedule or from a trigger; you can also start one yourself from the agent's page or from a conversation.

**Who can use it:** the **Member** role. Only agents of the **Workflow** type run this way; other agents start a conversation instead.

Video: Start an automation (2:07), https://katonicdocsmedia.blob.core.windows.net/docs/start-an-automation/b644a57dd564/index.html

## Prerequisites

- A published workflow agent. On the Agents page the **Workflow** chip filters the list to them. A card counts only your conversations with the agent, so a workflow agent you run rather than chat with reads **Not used yet**. A workflow agent's page shows **Automated workflow** under **What this agent can do**.

![The Agents page with the Workflow chip selected: the default agent's card at the top, then the type chips All, Conversational, Workflow, Coordinator and Used by me with the search box, and one card per workflow agent, each marked WORKFLOW with a Details link.](https://docs.katonic.ai/img/user-guide/work-room/start-an-automation/agents-workflow-filter.png)
*The Workflow chip on the Agents page filters the list to workflow agents, the only kind you can run as an automation. Select the chip again to clear the filter.*

## Start from the agent page

  
**Step 1. Open the agent**

Select **Details** on the agent's card, or follow **Start an automation** from the Automations page. The agent's page opens with **Back to agents**, **Run workflow** and its run history.

    ![A workflow agent's page: Back to agents, the agent's name and description, Run workflow, What this agent can do with the Automated workflow badge, the statistics Total runs, Success rate, Avg duration and Failed, the filters All, Failed and Slow, New run, and the list of past runs.](https://docs.katonic.ai/img/user-guide/work-room/start-an-automation/workflow-agent-page.png)
*A workflow agent's page ends in its run history. Run workflow at the top and New run above the list open the same dialog.*

  
**Step 2. Select Run workflow**

A dialog headed **Start this automation** opens with the agent's name and description.

    ![The Start this automation dialog over the agent's page: the agent's name and description, the Input box with a placeholder inviting any input or none, the note that results land in Artifacts, and the Cancel and Start buttons.](https://docs.katonic.ai/img/user-guide/work-room/start-an-automation/run-workflow-dialog.png)
*Run workflow opens the Start this automation dialog. With no defined inputs it offers one Input box; leave it blank to run with the agent's defaults, then select Start.*

  
**Step 3. Fill in the inputs**

What the dialog asks for depends on how the agent was built.

- With no defined inputs, it shows one **Input** box: add any input for this run, or leave it blank to use the agent's defaults. If the builder configured example inputs, chips under the box fill it with one click. Selecting a workflow agent's card on the Agents page opens the same dialog, with a required **Brief** box in place of **Input**.
- With defined inputs, it shows one field per input: a text box, a longer text area, a list to **Select…** from, knowledge source chips, or a file field. Required fields are marked with an asterisk, and **Start** stays unavailable until they are filled.
- A file field takes PDF, Excel, CSV, Word or text files, dropped in or added with **Add file(s)**. Each file is uploaded and its text extracted for the agent: **Uploading…**, **Extracting text…**, then **Text extracted**. A warning tells you when only the first part of a large file was extracted, or when no text could be extracted and the file will not reach the agent. **Upload failed** offers **Retry**; **Remove** (the cross) drops a file.

  
**Step 4. Start**

Select **Start**. It reads **Starting…** while the run is created, then the run's page opens so you can follow it; see [Automations and approvals](/user-guide/work-room/automations-and-approvals). **Cancel**, **Close** (the cross) or Escape closes the dialog without starting anything.

## Start from a conversation

You can also start a run from a conversation with the agent; see [Using agents](/user-guide/work-room/using-agents).

- If the builder defined inputs, the empty conversation shows a card headed **Workflow inputs · fill in and run** above the composer, with the same fields. Fill it in and select **Run workflow**: your inputs become your message in the thread, one line per field, and the run starts.
- If the builder defined no inputs, the message you send is the run's input.

### Follow the run in the thread

A status strip under the agent's message tracks the run:

- **Automation running**, with **N of M steps** and the name of the step in progress.
- **Stop** asks the run to stop; the strip reads **Stopping the automation** until it confirms, or **Couldn't stop the run** if the request failed.
- **Automation finished**, **Automation didn't finish** or **Automation stopped** when it ends.

Some runs also show a step tracker listing each step with its status (**Pending**, **Running**, **Awaiting approval**, **Done**, **Failed** or **Skipped**) and duration. Select a finished or failed step to read its output or error.

### Approve a paused step

When a step needs a person, the tracker shows **Human approval required**, the builder's prompt and, when the step carries data for review, **Entries under review**. If there is a deadline, a line says whether the step will be approved, declined or escalated automatically when time runs out.

- **Approve and continue** resumes the run; a notice confirms the automation is continuing.
- **Reject and stop** asks for a **Rejection reason (optional)**, then ends the run; a notice confirms it will stop.

If someone else decided first, the deadline passed, or you may not decide this step, the notice says so.

## Run history on the agent page

A workflow agent's page also shows **Total runs**, **Success rate**, **Avg duration** and **Failed**, then its runs filtered by **All (N)**, **Failed (N)** or **Slow**, which keeps the longest fifth. Each row shows the run's name, status, steps completed, duration and start time; select it to see its steps, or its prompt and output when it has no step records. **New run** opens the same dialog as **Run workflow**.

## Verify

The run's page opens after **Start**, and the run appears at the top of the Automations page and in the agent's run history. When it finishes, its results are listed in [Artifacts](/user-guide/work-room/artifacts).

## Troubleshooting

- **Couldn't start the run. Please try again.** in the dialog: the run was not created and your inputs are kept, so select **Start** again. From the Agents page the notice reads **Could not start run**; open the dialog again.
- **We couldn't load this automation's questions.**: the defined inputs could not be loaded; describe what you want in the **Brief** box instead.
- **Start** stays unavailable: a required field is empty, or a file is still uploading or being extracted; a file with no extracted text does not count.

## Related

- [Automations and approvals](/user-guide/work-room/automations-and-approvals)
- [Using agents](/user-guide/work-room/using-agents)
- [Artifacts](/user-guide/work-room/artifacts)

Source: https://docs.katonic.ai/user-guide/work-room/start-an-automation
