@Jessica Leighty that and n8n's memory limitations... 😆

I'll add some of my own and say first understanding how much of the process can and should be done on paper with concepting has helped my builds. I think a lot of us start with tossing nodes around in n8n and gradually sort through the process there. And I think that has some value in the beginning, seeing what works together and (mostly) what doesn't work. But just the exercise of thinking through the problem and working out all the dozens of potential ways to solve it can help narrow down to a clear path. Second, realizing that some questions are two-fold: "what kind of data do I have?", and "what kind of data do I NOT have?" I think we take for granted how much ancillary data we have as context for things. But going through the exercise of what data sets our workflows have access to and (more importantly) which they don't have access to can be a real eye-opener.

@Peter Hollis yes, this!

@Dan Quixote yes, having a scope of work laid out is very important! It sets and protects everyone's expectations for the project.

Can you share a screenshot of your workflow? I had a few iterations of a slack bot agent before I got mine working right. Your issue sounds like something I found resolution for, but I'll have to check my notes.

So, I have a question. I hosted my own instance inside a Docker container on a professional NAS I have. Even though it was connected to the internet, the n8n image was not visible TO the internet. Did some more digging and it seemed like I needed to create a domain and connect it for API access and whatnot, so there were some costs associated there (though I'm sure it's low). Are there ways to avoid this and still be able to host with zero additional costs?

@Peter Hollis thanks, Peter! I saw ngrok mentioned as a way to tunnel but was only recommendation for testing and not production. Would you use Cloudfare zero trust for production?