After answering a bunch of the same questions here and in a few other places, I put together answers to the ones that come up constantly:
Is n8n actually free?
Yes — self-hosted Community Edition has no caps on workflows/executions, you only pay server costs (~$4-10/mo on Hetzner/DigitalOcean). n8n Cloud starts around $20-24/mo if you don’t want to self-host.
Is it cheaper than Zapier?
Usually, especially at scale — n8n bills per workflow execution regardless of node count, Zapier bills per step. A 10-step workflow costs the same as a 2-step one in n8n.
Can I use ChatGPT/Claude with it?
Yes, both have dedicated nodes — a simple AI node for one-off generation, or an AI Agent node that reasons across steps and calls tools on its own.
I also cover self-host vs cloud, nodes/triggers, freelancing with it, and troubleshooting in more detail — happy to share if anyone wants the full writeup, or ask me anything here directly.
Good list, thanks. I want to add one thing about the cost part. (I build a hosted tool myself, so I am biased, but I also run my own server, so this is from real experience.)
The $4-10 for the server is correct. For me the real cost was different. I run 7 docker containers for my project: postgres, redis, queue worker, proxy and so on.
Every deploy I do by hand: ssh, git pull, docker compose build. Last week I found 1.8GB of old build cache filling my disk, and my proxy config on the server had 29 lines that were not in git. Nobody counts this part.
For my own project this is fine, I accept it. But if the workflow belongs to a client and it stops on saturday, someone still must answer the phone. That is not inside the $5.
You’re completely right, and this is exactly the gap that gets left out of most “DIY vs SaaS” cost comparisons — including mine. The $5-12/month API cost is real, but it’s only the compute bill, not the total cost of ownership.
What you’re describing — config drift, orphaned build cache, manual SSH deploys — is the actual tax of running your own infrastructure, and it doesn’t show up until months in. For a personal project it’s a fair trade. For anything with a client depending on uptime, you’re right that “someone answers the phone” is a real cost that doesn’t fit in a hosting bill line item.
I think the honest framing is: DIY wins on cost only if you’re also willing to own the ops burden, or you’re disciplined enough to keep it in something like a single docker-compose file in git with a proper CI deploy instead of manual SSH — which most of us don’t actually do once the thing is “working.” Appreciate you pushing back on this, it’s a blind spot worth naming explicitly.
1 Like
You named it exactly. I have the compose file in git, so I did teasy half. CI part I did not: my .gitlab-ci.yml is sitting in the repo with apipelines disabled, and I even wrote a spec for turning it back on. Never did it. Manual deploy works, so it is never the most urgent thing that day.
That is probably the honest version for most solo projects. The discipline is not hard, it just never wins against something that is already working.