Over this past year I have been working on building automated workflows with Make, Zapier, and n8n…in that order. The order is important…keep reading.
I started with Make (formerly Integromat) and was blown away with the power and simplicity of creating business-ready workflows. I built a social media posting machine that generated Facebook posts and Instagram reels daily based on current topics related to my businesses. However, there were a few integrations, mainly the Google Business API that forced me to look at Zapier. It was much easier in Zapier to set up an AI Agent that triggered when one of my businesses received a Google review to perform a well-thought-out reply. It worked perfectly…so why change?
Maximizing ROI
By this time I had several “active” scenarios and AI agents running in both Make and Zapier and I was reaching the limits of the 1st-tier paid level. Did I want to jump up to the next tier and pay even more monthly fees in addition to all the APIs I was paying for to run in the backend…Perplexity Pro, Anthropic/Claude, Open AI, DeepSeek, fal.ai, etc.??? No, there had to be a better way.
Enter n8n…
n8n was super easy to setup with their cloud account. It gave the power of Make and the easy integrations of Zapier all in one. I found the solution I was looking for…so I thought. n8n cloud gives you 5 active scenarios in their $25/month tier. That is fine for most, but I hate being constrained by a limited number of active scenarios. There are some workflow/agents I want to test that take days to run to get the settings just right. Even then it takes more time.
Hello self-hosting!!!
Yes! I am hosting n8n on a VM I created on my Google Cloud Console. The setup was ridiculously easier than I thought…I haven’t set up a virtual machine instance in years…man, has it come a long way!!! Not only was it easy to setup with an SSL, logs, and backups, it is CHEAP! I honestly think I could get it down to closer to $0/month if I didn’t create so many logs and backups! Oh well, old habits are hard to die. Once you have a server instance wiped away with years of work, you learn to BACKUP!!! lol
After I got my n8n self-hosting VM setup and running on my Google Cloud Account, and tested to the max, I started rewriting all my Zapier workflows, they call them “zaps”, with n8n. I got the Google Business API working perfectly after a few hiccups. Once that piece was done I quickly shut off the recurring billing for Zapier because it was the most expensive. I kept the free plan in case someone needs some help down the road. Once all my zaps were converted to n8n workflows, I moved on to rewriting all the Make scenarios. There were/are a lot of them. I started with all my active scenarios. The toughest part was converting all my code nodes. In Make I wrote Python to do any string manipulation or calculations. While n8n does support Python, JavaScript is the preferred scripting language. I quickly ran into trouble when I couldn’t include Python libraries into my self-hosted instance. There is a way to do it, but I was quickly approaching diminishing returns…and I found a BRILLIANT workaround.
Stock Analysis Agent
I have a simple buy/hold/sell stock agent that runs while the market is open and processes a list of stocks I give it. All the math was written in Python with some beautiful libraries that crunch stacks of data in milliseconds. The problem is, as mentioned earlier, n8n prefers JavaScript. Instead of banging my head against the wall trying to figure out how to get the needed libraries imported into my self-hosted n8n, or worse rewriting hundreds of lines of Python without the included helpful libraries, I did what any kid these days does, I went to ChatGPT! Actually, I used Claude from Anthropic. It is brilliant at code writing. I simply asked Claude to rewrite the code from Python to JavaScript and it quickly spit out the results. It thought, “Cool! Let me spend the next few hours working out the bugs and kinks and get it to work in JavaScript!” Guess what? It worked PERFECTLY the first time! No rewrites, no debugging, no hours tracing errors. I was thoroughly impressed! Needless to say, my little analysis stock program is working great now!
Keep following along as I post updates about my adventures in AI Workflow no/low Code Tools.