Getting Started with N8N: Hands-On Review

One afternoon, Yaniv, our beloved CEO, came to see our team with more enthusiasm than usual: someone had given him a demo of N8N and Yaniv had been impressed by the tool's UX, which in his words made chaining agents incredibly easy.

Intrigued, here I am writing a blog post about it.

What Is N8N?

N8N is a workflow automation platform. But what does that actually mean? It means that, like many other tools, both commercial and open-source (N8N is open-source), we're chasing the Holy Grail of knowledge workers since the digitalization of our economy:

• Automate repetitive tasks
• Model business processes
• Integrate these automations and models with external systems

Is N8N yet another overhyped promise (albeit with 54K stars on GitHub) or a tool that can realistically deliver on its claims? I'll try to build an intuition on this as I write this article!

Test Drive

Let's skip the blurb (interesting as it is) about the fair-code license and the (crucially important) fact that the tool is self-hostable: I'll let you check the repo for that. I'm too eager to try it out!

Apparently I just need NPM as a prerequisite. I created an n8n-demo folder and ran npx n8n; this should open a workflow editor in my browser, served locally.

I'm greeted by a quick questionnaire about my intended use of the tool and offered a "free activation key" for paid features (a bit odd for open-source software). I'm not sure what this questionnaire and the follow-up offer are about, but let's move on...

When the editor opens, I can choose between creating a workflow from scratch or using one of the templates. To have some fun and appreciate the UX from a newcomer's perspective, I'll start from scratch. I have no idea what I'm going to build; I'll let the tool guide me!

The tool seems to let you create workflows with the "step" as the fundamental building block, each with its own trigger. So we're dealing with something event-driven, which is interesting because these triggers are not constraints at all: from manual actions to elapsed time, file system changes on your machine, chat messages, and webhook calls, you're not limited by rigid activation conditions to trigger workflows, which is promising!

Ok, I've created my workflow entry point. Now what?

I agree with Yaniv: the UI is very intuitive and crystal clear. I understand what's happening at each step and feel guided, even though this is my first time using it! That's refreshing.

When I trigger my application via one or more entry points (triggers), I can do all sorts of things with my workflow (see screenshot above). Since this is the AI squad blog, I'll click on "Advanced AI" -> "Basic LLM Chain".

I have no idea what I'll find. Personally, I'd love to see what I can do with an SLM (Small Language Model) like Llama3.1:8B on my laptop...

... and it turns out it's extremely easy! I can even set parameters like Top K, temperature, etc. for my SLM.

Let's try structured data extraction: I'll feed in random news content and have the LLM output the 5Ws (who, what, when, where, how). I'll use a "Structured Output Parser" -> "Generate From JSON Example" for this.

{
  "input": "Saudi budget airline flyadeal is putting the finishing touches to a deal to order 10 Airbus A330neo jets in its first full-blown expansion into wide-body planes, as the kingdom pursues a surge of spending on aviation,",
  "WHO": "flyadeal, a Saudi budget airline (subsidiary of Saudi Arabian Airlines/Saudia)",
  "WHAT": "Finalizing a deal to order 10 Airbus A330neo wide-body aircraft",
  "WHERE": "Saudi Arabia",
  "WHEN": "Current/Ongoing (as of early 2024)",
  "HOW": "To execute their first major expansion into wide-body aircraft operations"
}

The idea is to pass a newspaper headline as input and receive the 5Ws as output. Let's run it!

The output I get is exactly what I expected:

{
  "output": {
    "input": "New Apple products expected this spring",
    "WHO": "Apple",
    "WHAT": "new iPhone SE 4, iPad 11 and more",
    "WHERE": "expected to be launched this spring",
    "WHEN": "spring",
    "HOW": "launch frenzy"
  }
}

What's Next?

Once you've discovered N8N, here are a few ideas to take things further within your organization:

• Set up a personalized instance for each team member so everyone can automate their work (with pre-configured API keys or local models)
• Explore the templates
• The possibilities are endless!

We're so enthusiastic that we're having lunch in front of the tool right now! Can't wait to share our experiments in future articles.