Getting started with workflow automation using n8n on gcube
on this page
Doing the same tasks every single day?
Opening report PDFs to pull out the key points, skimming the news every morning to share with your team, writing up action items after every meeting and posting them to Notion. Each one seems small, but together they eat up a surprising share of your day.
In this post, we show how to automate this kind of repetitive work by running n8n on gcube. We cover everything from setup to three real workflows, ready to use right away.
What is n8n?
n8n is an open-source workflow automation tool. Put simply, it lets you connect “when this happens, automatically do that” with no code or low code.
- Integrations with 400+ services, including Gmail, Slack, Notion and Google Sheets
- Built-in LLM integrations (GPT, Claude and more) for building AI processing pipelines
- Self-hostable, so your data never leaves your infrastructure
It’s similar to Make (formerly Integromat) or Zapier. The decisive difference is that you run it on your own server and use it without limits.
Running n8n on gcube
gcube deploys workloads as containers, so running n8n is very simple.
Container Settings
- Registry: Docker Hub (hub.docker.com)
- Container Image:
n8nio/n8n(official image, 220M+ pulls) - Container Port: 5678
The image has a built-in entrypoint (
tini -- /docker-entrypoint.sh), so n8n starts automatically when the container starts. No separate start command is needed.
After you deploy the workload in the gcube console, open port 5678 on the assigned endpoint and the n8n dashboard appears.
A gcube workload runs 24/7 as long as it’s deployed. The key point is that your automations keep running even when your local PC is off.
Scenario 1. Upload a PDF → auto-summarize → send to Slack
Pain point: Opening and reading every PDF (reports, proposals, meeting materials) one by one is tedious.
Node chain: On form submission → Extract from File → Message a model → Send a message
Building the workflow (PDF summary)
① Add an On form submission node
Open a new workflow, click the + button, type “n8n form” in the search box and select it. n8n automatically generates a web form URL. Share this URL with your team, and anyone can upload a PDF from their browser.
Settings:
- Form Title: PDF summary request
- Field: File (File type, Required On)
- Field: Document title (Text)
② Add an Extract from File node
Click the + on the right of the On form submission node to connect it.
- Operation: Extract from PDF
- Input Binary Field: File (enter it exactly as the form label)
After running the node, check that the PDF body was extracted into the text field on the Output tab.
If you insert an Edit Fields node in “Set” mode before Extract from File, the binary data is lost. If you really need one, use “Append” mode.
③ Add a Message a model node (OpenAI)
Type “Message a model” in the search box and select it.
- Credential: connect your OpenAI API key
- Model: gpt-4o, or gpt-4o-mini to save cost
- Messages → enter the prompt below as the User message
Summarize the following document in 3 lines. Be sure to include any key figures or dates.
{{ $json.text }}
You can see the expression {{ $json.text }} replaced with the actual PDF text in the preview.
If you already run Ollama on gcube, you can call your local LLM endpoint with an HTTP Request node instead of this node. That way there are no external API costs.
④ Add a Send a message node (Slack)
Type “Send a message” in the search box and select the Slack entry.
- Credential: connect Slack OAuth or a bot token
- Send Message To: choose the channel to post to (e.g. #pdf-summaries)
- Message Text:
📄 PDF summary ready
Document: {{ $('On form submission').item.json['Document title'] }}
{{ $json.message.content[0].text }}
Check that both expressions resolve correctly in the preview.
⑤ Activate the workflow
Click the Publish button at the top to activate it. Saving alone doesn’t turn on the trigger. Once it’s active, bookmark the generated form URL or pin it in your team channel so it’s ready to use.
Scenario 2. Collect AI news every morning → summary report → email
Pain point: Keeping up with the industry means visiting several sites every morning.
Node chain: Schedule Trigger → RSS Feed Read ×2 → Merge → Limit → Aggregate → Message a model → Send a message
Building the workflow (news briefing)
① Add a Schedule Trigger node
- Trigger Interval: Days
- Days Between Triggers: 1
- Trigger at Hour: 8 / Trigger at Minute: 0
n8n’s default time zone is America/New_York. Click the gear icon at the top right of the workflow and set Timezone to your local time zone (for example, Asia/Seoul), then save. If you don’t, the workflow runs at 8 AM New York time instead.
② Add two RSS Feed Read nodes in parallel
Click + on the Schedule Trigger to add the first RSS Feed Read node. Click + on the Schedule Trigger again to add a second node in parallel.
- Node 1 URL:
https://techcrunch.com/feed/ - Node 2 URL:
https://www.technologyreview.com/feed/
To add more sources, add more parallel nodes the same way and connect them to the Merge inputs.
③ Combine sources with a Merge node
- Mode: Append
- Connect the outputs of the two RSS nodes to input 1 and input 2
When it runs, 20 items from TechCrunch + 10 from MIT Technology Review = 30 items are combined into a single list.
④ Add a Limit node
- Keep: First Items
- Max Items: 10
Passing all 30 items to the LLM drives up token costs, so we cap it at 10. Note that Append lists TechCrunch’s 20 items first, so this keeps only TechCrunch items. To mix both sources, add a Sort node (by publish date, descending) before the Limit node.
⑤ Add an Aggregate node
- Aggregate: All Item Data (Into a Single List)
This bundles the 10 separate items into a single array (the data field) and passes it to the LLM in one go. In the execution result, 10 items are reduced to 1 item containing the data array.
⑥ Add a Message a model node (OpenAI)
- Credential: connect your OpenAI API key
- Model: gpt-4o or gpt-4o-mini
- Messages → enter the prompt below as the User message
Below is today's list of collected news.
Pick only the items about AI, cloud and GPUs,
and format them like this:
📌 Title
One-line summary
🔗 Link
{{ JSON.stringify($json.data) }}
You can see the expression {{ JSON.stringify($json.data) }} replaced with the actual news JSON in the preview.
⑦ Add a Send a message node (Gmail)
Type “Send a message” in the search box and select the Gmail entry.
- Credential: connect Google OAuth
- To: enter the recipient email address
- Subject:
[Daily Briefing] {{ $now.format('yyyy-MM-dd') }} AI/Tech News(resolves correctly in the preview, e.g. 2026-06-18) - Email Type: Text (keeps line breaks; in HTML mode they collapse into a single paragraph)
- Message:
{{ $json.message.content[0].text }}
---
This email was sent automatically from gcube.
⑧ Activate the workflow
Click the Publish button at the top to activate it. To test right away, click Execute step on the Schedule Trigger node.
Scenario 3. Meeting notes text → extract action items → save to Notion
Pain point: Writing up and posting action items after every meeting is a chore.
Node chain: Webhook → Message a model → Notion → Send a message
Building the workflow (meeting notes)
① Add a Webhook node
In the trigger selection panel, choose On webhook call.
- HTTP Method: POST
- Path: meeting-summary
- Respond: When Last Node Finishes
When you add the node, a test URL and a production URL are generated automatically.
Test: https://[gcube-endpoint]/webhook-test/meeting-summary
Production: https://[gcube-endpoint]/webhook/meeting-summary
Click Listen for test event to put it in listening mode, then send a POST request with the JSON structure below.
{
"title": "Weekly marketing meeting",
"date": "2026-06-18",
"body": "Attendees: ...\nAction items: ..."
}
When it’s received, you can see the body.title / body.date / body.body fields parsed on the Output tab.
gcube service URLs are tied to your browser login session. Calling them directly with curl from an external terminal won’t connect. Call them with JavaScript
fetchfrom the same browser session, or from a server on the same network.
② Add a Message a model node (OpenAI)
- Credential: connect your OpenAI API key
- Response Format: Text (the prompt asks for a JSON array, and JSON Object mode only accepts a top-level object)
- Messages → enter the prompt below as the User message
Extract only the action items from the meeting notes below.
Respond as a JSON array. Do not include any other text.
Format:
[
{ "owner": "name", "task": "to-do", "due": "date or TBD" }
]
Meeting notes:
{{ $json.body.body }}
③ Parse with a Code node (optional)
If the OpenAI response comes back as a string, parse it with the code below.
const raw = $input.item.json.message.content[0].text;
const items = JSON.parse(raw);
return items.map(item => ({ json: item }));
④ Add a Notion node
First, set up the n8n integration in Notion.
- Notion → Settings & Members → Connections → Develop or manage integrations
- Create a new integration and copy the Internal Integration Token
- On the Notion database page you’ll use, click ... at the top right → Add connections → add the integration
- Enter the token in the n8n Notion node
Node settings:
- Resource: Page / Operation: Create
- Database: select the connected Notion database
- Title:
{{ $('Webhook').item.json.body.title }} action items - {{ $('Webhook').item.json.body.date }} - Properties: map to the Owner / Task / Due columns
You must share the integration with each database you use, one by one. If a database doesn’t show up in the dropdown, this step was most likely skipped.
⑤ Add a Send a message node (Slack)
- Credential: connect Slack OAuth or a bot token
- Send Message To: choose the channel for notifications
- Message:
✅ Meeting notes processed
Meeting: {{ $('Webhook').item.json.body.title }}
Action items have been saved to Notion.
⑥ Activate the workflow
After you publish, post the production webhook URL in a shared team document. Anyone can then paste meeting notes and organize them in Notion with a single POST request.
Setting up credentials
All three scenarios need the credentials below to connect to external services. n8n stores credentials encrypted; set them up by clicking the Credential field on each node.
- OpenAI → API key (scenarios 1, 2 and 3)
- Slack → OAuth or bot token (scenarios 1 and 3)
- Gmail → Google OAuth (scenario 2)
- Notion → Internal Integration Token (scenario 3)
Once the credentials are set, click Publish at the top of each workflow and it’s fully up and running.
Wrapping up
All three scenarios follow the same pattern:
Collect → process with an LLM → deliver
Once you’ve got this pattern down, you can automate most repetitive work with n8n. You can extend it to detecting Slack keywords, alerting on Google Sheets changes, or even automating contract reviews.
Running it on gcube keeps it going 24 hours a day regardless of whether your PC is on, and connecting a local LLM lets you process sensitive data without sending it outside.
gcube is a GPU cloud platform that pools heterogeneous GPUs and runs workloads in containers. You can run workloads that don’t need a GPU, like n8n, and operate AI inference servers or LLM services on the same infrastructure.