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

# AI Automations

> Build workflows on a canvas - triggers, agent steps, conditions, approval gates, and outputs - and let them run on a schedule, an alert, or a webhook.

The **Automations** designer under **AI** → **Automations** is where you build repeatable AI workflows as a node graph: a trigger starts the workflow, agent steps do the work, conditions route on results, approval gates pause for an administrator, and an output delivers the result. The page has three panels - the **Workflows** list, the canvas, and the **Inspector** for the selected node.

## Anatomy of a workflow

| Node          | Role                                                                                        |
| ------------- | ------------------------------------------------------------------------------------------- |
| **Trigger**   | Starts the workflow                                                                         |
| **Agent**     | Investigates or takes action - a goal prompt plus a model, tool profile, and autonomy level |
| **Condition** | Routes by an earlier result - match on the previous step's outcome, status, or summary      |
| **Approval**  | Waits for an administrator before continuing                                                |
| **Output**    | Completes the workflow and posts a notification message                                     |

Every workflow needs a trigger. The trigger kind decides when it runs:

* **Manual run** - only when someone presses **Test run**.
* **Schedule** - a cron expression with a timezone.
* **Alert fires** - bound to one of your [alert rules](/guides/alerts).
* **Webhook** - an external system calls in.

## How a run executes

Ankra walks the graph node by node. An agent step opens an agentic session with the step's goal prompt and waits for it to finish; a condition picks the branch whose comparison matches; an output posts its notification and completes the run. Recent runs are listed per workflow in the Inspector.

When a run reaches an **Approval** node, it pauses and files a ticket on the [AI Board](/platform/ai-board) directly into **Awaiting approval**. Approving - from the Inbox or the ticket - resumes the run; denying cancels it. The run's outcome settles the ticket when it finishes.

<Note>
  Approval gates are how an automation stays safe on a schedule: the workflow can investigate and prepare autonomously overnight, but nothing irreversible happens until a person signs off.
</Note>

## Templates

**Start from a template** in the Workflows list seeds a ready-made graph you can adapt:

* **Nightly cluster fix plans** - scheduled sweep that drafts fixes and parks them for approval.
* **Daily insight digest** - collects the day's findings into one notification.
* **On alert: diagnose** - triggered by an alert rule, investigates, and reports.

## Editing and permissions

Workflows are **Enabled** or **Paused**, and the command bar shows an **Unsaved** chip until you press **Save**. **Test run** is available once an enabled workflow is saved. Only organisation admins can edit; everyone else sees the designer read-only. Deleting a workflow lives in the Inspector's danger zone.

Automations dispatch agent sessions per step, which makes them the graph-shaped complement to [AI Agents](/platform/ai-agents) - standing agents with a single trigger and goal. Reach for an agent when one prompt on one trigger is enough, and an automation when the work has stages, branching, or an approval in the middle.

## Availability

Automations are enabled per organisation, separately from the rest of the AI section. If **Automations** is missing from the AI sidebar, ask your Ankra contact to enable it.