n8n's licence is free and its true cost is engineering hours: hosting, upgrades, credential management and debugging workflows nobody documented. If you have a technical owner, that trade is excellent. If you don't, the realistic alternatives are Make for complex branching automation, Zapier for simple plumbing, Lindy or Gumloop for agent building without infrastructure, and a pre-built agentic platform like Operater if you want the function covered rather than the workflow designed.
n8n is genuinely excellent software. Self-hostable, a huge node catalogue, capable AI nodes, and no per-task billing to make you flinch. For a technical team it is close to the best value in the category.
It is also the tool most often recommended to people who should not be using it. The recommendation usually comes from someone technical, for whom the cost is invisible, to someone who is about to discover it.
The real cost of self-hosting
The licence is free. The rest is not.
- Hosting and uptime. Something has to run it, and something has to notice when it stops.
- Upgrades. Breaking changes exist. Someone reads the release notes.
- Credential management. API keys across a dozen services, rotated, stored safely.
- Design time. Every workflow is built by hand, and this is the big one.
- Debugging. A workflow that silently stopped three weeks ago is a genuinely unpleasant afternoon.
- Institutional memory. The person who built it leaves. The flows remain, undocumented.
Call it a few hours a month steady-state, plus a serious chunk during setup. At a founder's effective hourly rate that is not free. It is one of the more expensive things in the stack.
The question that decides it
Not "can we handle it?": most teams can. The question is: is automation part of anyone's actual job description?
If yes, n8n is probably right and you should stop reading. If it is a thing people do between other things, self-hosted infrastructure becomes the project that is 80% finished for a year.
The alternatives
| Option | You still design? | You still host? | Suits |
|---|---|---|---|
| n8n Cloud | Yes | No | Teams who want n8n without servers |
| Make | Yes | No | Complex branching automation, tight budget |
| Zapier | Yes | No | Simple, deterministic plumbing |
| Lindy | Yes | No | Agent building without infrastructure |
| Gumloop | Yes | No | Marketing-weighted workflows |
| Operater | No | No | Sales and marketing covered without design |
n8n Cloud: the smallest change
If you like n8n and only want the servers to go away, this is the obvious move. Be clear about what it fixes: hosting and upgrades. It does not fix design time, which is the larger cost for most teams.
Make: the value option
Make handles genuinely complex branching scenarios at a lower price than Zapier for equivalent volume, with a visual editor most non-technical operators can learn in an afternoon. Still fixed-path automation, so it inherits the exception problem described in Zapier vs AI agents.
Zapier: the simple option
Fastest to a working flow, most expensive at volume, and structurally limited to deterministic paths. For plumbing between two systems, that is exactly right and nothing else is faster.
Lindy and Gumloop: building without infrastructure
Both give you genuine agent capability with a friendlier builder and no servers. Lindy is broader; Gumloop skews to marketing. You are still the architect, which is fine if you want to be.
Operater: removing the design step
The different answer: agents that arrive already knowing a business function. Nothing to design, nothing to host. Connect tools, describe the business once, and sales and marketing agents run. Free for 150 actions a month, then $39 with unlimited seats.
The trade is control. n8n will always do more arbitrary things than a pre-built agent. If your workflow is genuinely unusual and someone wants to own it, keep n8n.
A decision rule
Name the person who will maintain this in six months. If you cannot name them, or the name is "me, when things calm down," do not self-host. That sentence has never once been true.
Key takeaways
- "Free" self-hosted software costs whatever your team's hours are worth, forever.
- The decision is not technical capability: it's whether automation is anyone's actual job.
- n8n Cloud removes hosting but not the design work, which is the larger cost.
- If you can't name who will maintain it in six months, don't self-host.
- Pre-built agents remove the design step entirely; that's the whole trade for less control.