Build Your First Workflow

Create a workflow that receives a webhook, analyses the payload with AI, and posts the result to Slack.

This tutorial builds a working automation end to end: a webhook receives text, an AI node analyses it, and a Slack node posts the result. You will run it from the editor first, then publish it and call it over HTTP.

Before you start

A Fabric account
An AI provider configured under AI configuration
A Slack connection, if you want the last step — the workflow works without it
curl and openssl, for the HTTP call at the end

Step 1: Create the workflow

Open Workflows

Click Workflows in the sidebar, then New Workflow.

Name it

The workflow is created straight away as Untitled Workflow and opens on an empty canvas. Click the name in the header and change it to My First Workflow.

Step 2: Add the trigger

Add a trigger node

Click Add a Step on the canvas. The right-hand panel opens on its Actions tab; choose Trigger.

Set the trigger type

Click the node and set Trigger Type to Webhook. The URL and its signing secret are generated when you publish.

Step 3: Analyse the payload with AI

Add a Generate Text node

Click + in the toolbar. An empty action appears to the right and the Actions tab opens; choose Generate Text from the AI group.

Adding a node does not connect it. Drag from the handle on the right of the trigger to the handle on the left of the new node — an unconnected node never runs.

Write the prompt

Open the node and set its prompt:

Analyse the following text and provide:
1. A one-sentence summary
2. Key topics mentioned
3. Sentiment (positive, negative, or neutral)

Text: {{body}}

Every top-level field of the webhook's JSON payload becomes a variable of the same name, so a request sending {"body": "..."} gives you {{body}}. To use another node's output instead, reference it by that node's label: {{Generate Text.text}}.

The node runs on your workspace's configured model; it has no model picker of its own. (Generate Image does — its model is part of the node's settings.)

Step 4: Post the result to Slack

Skip this step if you have not connected Slack — the workflow still runs, and you can read the AI output in the execution log.

Add the node

Click + again, choose Send Slack Message, and connect it to the AI node the same way.

Connect Slack

If Slack is not connected yet, open Integrations in the sidebar, connect Slack, and come back. See the Slack integration guide.

Write the message

Channel: #general, or any channel your Slack app is in
Message:

New analysis complete

{{Generate Text.text}}

Generate Text is the AI node's label and text is the field it outputs. Rename the node and you have to update the reference to match.

Step 5: Save, then run it from the editor

Press Save, then Run. The execution panel shows each node in turn with its input, output and duration.

The same executor runs both paths, so a node that works here works over the webhook — with two differences worth knowing before you read the results. An editor run sends no webhook payload, so {{body}} resolves to nothing and the AI node sees a prompt ending in Text:. And publishing uses your saved graph, not what is on screen: nothing autosaves, so save before you publish or you will publish an empty workflow.

Step 6: Publish and call it over HTTP

Publish

Click Publish, tick Enable Webhook Trigger, and add a changelog line if you want one. Ticking that box is what arms the endpoint: it sets the workflow's trigger type to Webhook and mints the signing secret. Publish without it and calls are refused, however published the workflow looks.

Copy the URL and the secret

The dialog shows the webhook URL and the webhook secret. Copy both now: while the workflow stays published, nothing shows the secret again.

Sign a request and send it

The endpoint refuses unauthenticated requests. Sign the exact bytes you are sending with the secret, and pass the hash in X-Workflow-Signature:

URL="https://…/api/workflows/trigger/<workflowId>"
SECRET="<the secret from the publish dialog>"
BODY='{"body": "Our team launched the new dashboard this week. Feedback has been positive, with a 40% increase in daily active users. Mobile performance still needs attention."}'

SIGNATURE=$(printf '%s' "$BODY" | openssl dgst -sha256 -hmac "$SECRET" -r | cut -d' ' -f1)

curl -X POST "$URL" \
  -H "Content-Type: application/json" \
  -H "X-Workflow-Signature: sha256=$SIGNATURE" \
  -d "$BODY"

A successful call returns an executionId.

Watch the run

Open the workflow's execution history and find the run with that executionId. Each node shows its status; click one to see what went in and what came out. If you wired Slack, the message is in your channel.

Step 7: Extend it

Branch on the result

Add a Condition node after the AI node. Its expression is evaluated against earlier nodes' outputs, and the node gets a true and a false edge you can wire to different destinations:

{{Generate Text.text}}.includes("negative")

Send true to #urgent-alerts and false to #general-updates.

Run it on a schedule instead

Change the trigger node's type to Schedule and give it a cron expression such as 0 9 * * * for every morning at 09:00. The schedule starts when you publish; pausing the workflow from its card menu stops it.

Let AI draft the next one

Instead of building by hand, describe what you want:

Create a workflow that receives customer feedback via webhook,
analyses sentiment with AI, creates a Linear issue for negative
feedback, and posts a summary to Slack.

Habits worth keeping

Know the three reference forms. {{Node Label.field}} reads an earlier node's output, {{fieldName}} reads a field of the trigger payload, and {{$nodeId.field}} does the same as the first but survives renaming the node.

Run after each node you add. Catching a bad reference with two nodes on the canvas is much easier than with eight.

Rename nodes to what they do. Analyse Sentiment beats Generate Text when a graph has three AI nodes — and the label is what you reference.

Troubleshooting

The call returns 403. Either the workflow is not live — a draft, paused or archived — or its trigger type is not Webhook. The response body says which. The second case is the common one: republish and tick Enable Webhook Trigger.

The call returns 401. The signature does not match. Sign the exact bytes you send — re-formatting the JSON after signing changes the hash — and prefix the hex digest with sha256=.

The AI node returns nothing. Check that an AI provider is configured under Settings → AI Providers, then open the node's entry in the execution log and confirm the prompt arrived with {{body}} filled in.

Slack did not receive the message. Confirm the Slack app is in the target channel and read the Slack node's output in the execution log — a failed call reports the reason there.