Self-Hosting n8n Automations
Yesterday our web server's config reload told us it had succeeded. It hadn't. The proxy container had the old config bound by the file's identity, not its name, so my edit went in, the reload said "OK," and the server kept right on serving the old rules. No error. No warning. Just a lie.
That's self-hosting in one story. I still choose it, most days. But you should know what you're choosing.
What n8n is, in case you don't know
Many of you already know this, but in case you don't: n8n is a workflow tool. Something happens -- a form gets filled out, a webhook fires, a schedule hits -- and n8n runs a chain of actions you wired together with a visual editor. It ships with hundreds of ready connectors: Google Sheets, Slack, Twilio, Stripe, HubSpot, Salesforce, you name it. The source code is on GitHub, and you can read every line of it.
Here's the part most people get wrong: n8n is not open source in the usual sense. It's "fair-code" under their Sustainable Use License. You can self-host it free for your own business, forever. What you can't do is resell it as a service to other people. For a small business using it internally, that distinction costs you nothing. But read the license if hosting for clients is your plan.
Or you can skip all of that and use n8n Cloud: the Starter plan is €24 a month for 2,500 workflow runs. Fine price. Metered, though -- every run counts, and the meter moves up faster than you'd guess once a phone line is involved.
Why a voice agent cares
A voice agent that only talks is a toy. The money is in what the call does.
Our AI receptionist answers a call, has a real conversation, and captures what matters -- name, number, intent. Then it needs to hand that somewhere. n8n's webhook trigger is the door: when the call ends, the agent posts a small packet of JSON to a URL you control, and the workflow takes over. The caller hears "we'll get right back to you," and meanwhile a machine is already doing the getting-back.
What does that look like concretely? A few we've sketched for real businesses:
- Missed call comes in after hours. The agent captures the number and reason. n8n creates the lead in the CRM, texts the caller a booking link, and posts a heads-up in the owner's Slack.
- A plumbing customer describes an emergency at 11 PM. The agent triages it, and n8n pages the on-call tech by SMS instead of waiting for morning.
- Job wraps up. The agent logs it, and n8n schedules the invoice, then a review request two days later.
- Someone wants to pay a deposit over the phone. The agent takes the details, and n8n generates a Stripe payment link and texts it before the call even ends.
None of these workflows are exotic. That's the point -- the phone was always the bottleneck, not the automation.
The advantages of running it yourself
The cost math changes with volume. €24 a month is nothing until your agents make 2,500 calls' worth of triggers a month; then it's €60, then more. A small VPS runs n8n for the price of a few lunches and doesn't count executions. If voice is your front door, you will blow past cloud tiers fast.
Your data stays yours. Call details, customer phone numbers, deal notes -- everything a voice agent captures flows through your automation layer. Self-hosted, that data lands on a machine you control and goes nowhere else. For anyone in a sensitive line of work (legal, medical, financial), that's not a nice-to-have. It's the requirement.
No vendor outage takes your phones down. When a cloud automation platform has a bad day, your workflows just... stop, and you find out from customers. Self-hosted, your ceiling is your own box.
And if you already run your own stack -- and we do -- n8n slots in next to everything else. Same box, same backups, same rules.
The disadvantages, honestly
You are the operations team now. Updates, backups, monitoring, disk space -- all yours. Nobody ships a fix to your server while you sleep. The n8n self-hosting docs are genuinely good, and Docker makes the install simple, but "simple install" and "nobody's pager goes off at 2 AM" are different products.
Security is on you too. A webhook that triggers workflows is a door into your business; a self-hosted one is a door you left the security decisions for -- authentication, HTTPS, what that endpoint is allowed to touch. Get that wrong and your automation layer becomes someone else's entry point.
Upgrades can break things. Community nodes lag behind core updates. A workflow that ran fine for months can fail after an update for reasons that take an evening to find. Cloud users get migrated; you get to be the migration.
And remember that config reload from the top of this post? That wasn't even n8n -- that was ordinary self-hosted infrastructure, and it still cost us an evening. Multiply that pattern across everything you host, and you understand the real price. Self-hosting doesn't remove problems. It just changes whose name is on them.
Where we land
Pretty sure most small businesses should start on n8n Cloud, prove the workflows matter, and move to self-hosting when execution counts or data rules force the question. We went the other direction -- we self-host first, always have -- because the voice agents we build are touching real customer conversations, and we want that data on machines we control. Both roads work. Pick the one whose bad days you can live with.
If you're wondering what a voice agent could trigger in your business, that's literally the conversation we love having. Tell us what your phones do today, and we'll sketch the workflow with you.