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

# Configure AI automation

> Build a workflow on the canvas, distinguish publishing from activation, and monitor or pause its deployment.

Automations run a saved workflow when a matching event occurs. Building or publishing a workflow does not turn it on: a deployment must be activated separately.

## Before you start

You need an Owner or Admin role, and Evelin must be available in your workspace. Decide which events should start the workflow and what changes it may make. An automation can consume AI credits even when it does not send a reply.

## Build and save a draft

1. Open **Evelin AI → Automations** and select **New automation**.
2. Give the workflow a recognizable name.
3. Use the canvas to inspect the starting trigger and existing steps. Select a step to configure it.
4. Use **Blocks** to add a **Branch**, **Wait**, **AI task**, **Update ticket**, or **Call service** when needed. Connect the steps in the order they should run.
5. In the selected block's configuration, choose **Apply changes**. This applies the block to the local draft; it is not the same as saving the automation.
6. Select **Create** for a new automation, or **Save** for an existing draft.

Configure required agents, destinations, or other values before publishing. A connector step needs an appropriate active connection; connecting Linear to Evelin is a separate workflow from configuring an automation webhook or Telegram destination.

## Understand the mode

The editor determines the mode from the actions in the workflow:

| Mode | Meaning |
| - | - |
| Observe | No write actions are included. AI can still generate internal results and use credits. |
| Assist | The workflow includes permitted write actions, but not an automatic customer-reply step. Do not assume every action waits for a teammate's approval. |
| Autopilot | The workflow includes sending a customer reply automatically. |

The allowed actions and channel scope matter as much as the mode label. Do not assume an email workflow also supports sending WhatsApp or Live Chat replies.

## Publish and activate

1. Review the graph and select **Publish**. This creates the reviewed version; it does not start runs.
2. Select **Create inactive deployment**.
3. Select **Activate** and review its mode, channel scope, hourly AI budget, operation ceiling, and allowed actions.
4. If it can send customer email, read and confirm the additional customer-messaging acknowledgment.
5. Confirm activation only when those actions should begin processing eligible events.

Creating a new draft does not replace a version that is already active. Check the deployment state as well as the draft or published version label.

## Monitor, pause, and revise

Open **Evelin AI → Monitor** to inspect runs and issues. If a workflow does not run, check its deployment state, trigger conditions, required connections, and reported budget or permission errors.

Use **Pause** on an active deployment when it should stop processing new work. Pausing does not reverse completed changes or recall emails already sent.

For a published workflow, select **Create new draft**, make and save the changes, then publish and review the new deployment before activation. If you see **Review & activate**, inspect the changed authority before reactivating.

See [Understand AI credits](/guides/understand-ai-credits) for balance and usage guidance.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.