We use privacy-friendly analytics to understand how the site is used. You can opt out any time on our Do Not Sell or Share page. See our Privacy Policy.
For most AI agent client work, put the client on n8n Cloud (starts at $24/mo) and self-host only when the client needs data control, high execution volume, or custom code the Cloud sandbox blocks. The deciding factor is rarely price. It's who owns the risk when a workflow breaks at 2am. Self-hosting saves real money at scale, but a $5 Hetzner box you babysit for free is not a win when you could bill for a managed setup instead.
The n8n cloud vs self-hosted question is one of the most common from freelancers taking AI agent jobs, and the answer changes depending on whether you're the one holding the pager. AI Agent Development is the single biggest hiring category on DevSnipe, with 425 jobs posted in the last 30 days across the 5 platforms we track. A large share of those builds run on n8n. So the hosting call comes up on nearly every contract.
If you're still deciding between n8n and another platform before you get to hosting, read our Make.com vs n8n comparison for freelance client work first. This guide is the hosting follow-on: once you've picked n8n, this is how you decide where to run it.
n8n Cloud is the hosted version n8n runs for you, billed monthly with execution caps. Self-hosted is the same software (the Community Edition is free and source-available under n8n's fair-code Sustainable Use License) that you run on your own server, with unlimited executions and full control. The code is nearly identical. What differs is who patches it, who backs it up, and who gets the call when it dies.
One licensing note before you self-host for a client. n8n is source-available, not OSI open source, so under the Sustainable Use License you can run it for your own or a client's internal business use, but you can't resell it as your own hosted service without a commercial license. Running one client's workflows on their own server is fine. Spinning up a multi-tenant n8n product to sell is not.
Here's how the Cloud plans price out, per n8n's pricing page:
The free Community Edition has no execution limit at all. The catch is you supply everything else: a server, a database, backups, SSL, monitoring, and your own time keeping it alive.
One workflow execution on Cloud counts as one run of the whole workflow no matter how many nodes fire. That matters for AI agents, because an agent that loops through tool calls can chew through executions fast. A chatbot handling 100 conversations a day, each triggering one workflow run, is 3,000 executions a month. That's already past the Starter cap.
Here's the side-by-side that actually drives the decision on a client build. Cloud wins on speed to ship and zero maintenance. Self-hosted wins on cost at scale, data control, and unrestricted code execution.
| Factor | n8n Cloud | Self-Hosted |
|---|---|---|
| Starting cost | $24/mo (Starter) | ~$5-6/mo VPS, free software |
| Execution limit | Capped per plan | Unlimited |
| Setup time | Minutes | Hours to days |
| Maintenance | n8n handles it | You handle it |
| Custom code / npm packages | Restricted sandbox | Full access |
| Data residency | n8n's infra | Your server, your region |
| Backups | Automatic | Your job |
| Who fixes a 2am outage | n8n | You |
| Best for | Fast client delivery, low volume | High volume, data-sensitive, custom |
A production self-hosted setup is not the $5 number people quote. A bare VPS running n8n on SQLite starts around $5-6/mo on a provider like Hetzner (prices current as of 2026). Add what a client build actually needs and the real number climbs:
All in, you're realistically at $24-30/mo on Hetzner, per this cost breakdown from openhosst. At that point the raw hosting cost is a wash with Cloud. The difference is entirely in the labor and the control.
Use n8n Cloud when the client's volume fits inside a plan tier, they don't have hard data-residency rules, and you want to hand off something you're not on the hook to maintain. For maybe 6 out of 10 client builds, Cloud is the right call. It ships in an afternoon and the client owns the account.
The strongest argument for Cloud on client work is liability. When you self-host on infrastructure you set up, you implicitly own the uptime. Alex, who runs DevSnipe, puts it plainly: "The cheapest VPS in the world is expensive the first time a client's lead-capture agent goes down over a weekend and you're the only one who knows how to restart it." A managed platform moves that risk off your plate.
Cloud also makes the billing clean. The client pays n8n directly with their own card. You're not fronting server costs, marking them up, or explaining a hosting line item on your invoice. You build the workflow, you get paid for the build, and you walk away with an optional maintenance retainer. If you want to structure that recurring income, see our guide on landing AI automation retainer clients.
One more Cloud advantage that's easy to miss: the client can log in and see their own workflows. Non-technical clients like a dashboard they control. It reduces the "can you send me a screenshot" messages by a lot.
DevSnipe sends you AI automation jobs from Upwork, Skool, and more before most people see them.
Self-host when the client needs data to stay on their own infrastructure, when execution volume would blow past Cloud pricing, or when the agent needs custom npm packages or code the Cloud sandbox won't run. These are real constraints, not preferences. If any one of them is a hard requirement, Cloud is off the table.
Data residency is the most common trigger. Healthcare, legal, and finance clients often can't send workflow data through a third-party host, or they have a compliance officer who says no. Self-hosting on the client's own cloud account (their AWS, their Hetzner, their region) keeps the data where their contracts require. This is where a lot of higher-paying AI agent development jobs live, because the constraint filters out freelancers who only know Cloud.
Volume is the second trigger. An agent doing tens of thousands of executions a month gets expensive on Cloud fast. At 40,000 executions you're looking at the $800 Business tier, versus a self-hosted box handling the same load for the cost of the server. The math flips hard past a certain scale.
The third trigger is code. n8n Cloud runs the Code node in a restricted sandbox. If the agent needs a specific npm library, shell access, or a long-running process, you need self-hosted. This shows up a lot in agent work that touches internal tools or unusual APIs.
If you do self-host for a client, do it on their infrastructure, not yours, and charge for the setup and the ongoing management. Never run a client's production workflow on a personal server you're maintaining for free. That's how you end up doing unpaid ops work for years.
There are three clean ways to handle billing when you're hosting n8n for client work. Pick one and put it in writing.
The retainer model is the best for your income. It turns a one-off build into recurring revenue, and the management fee is where self-hosting pays off for you, not the client. The trap to avoid is quoting a self-hosted build with no ongoing fee, because then you're on the hook for maintenance forever with nothing coming in.
Freelancers on n8n work command solid rates. Our June 2026 rates report breaks down what automation builders are actually charging, and hosting management is one of the cleaner ways to add recurring line items on top of project fees. For pricing the build itself, see how to price AI agent development projects.
Whatever you pick, write the hosting arrangement into the contract. Who owns the account. Who pays the bill. What happens to the workflows if the client stops paying you. Vague hosting terms cause more freelancer disputes than the actual build ever does.
Cloud is more reliable out of the box because n8n handles the infrastructure, but self-hosted can match it if you set up PostgreSQL, backups, and error handling correctly. The reliability gap is real for a default self-hosted install. It mostly disappears with a proper production config. The handoff is where most client builds actually fail, regardless of hosting.
A self-hosted n8n running on SQLite with no backups is a time bomb. Move to PostgreSQL, set automated backups, and add a queue mode if the client has real volume. Skip that and the first server reboot can corrupt the database. Cloud gives you all of this by default, which is a chunk of what the $24/mo actually buys.
Regardless of where it runs, build workflows that fail loudly and recover cleanly. That means retries, error branches, and alerts the client actually sees. We cover this in depth in building client-ready n8n workflows. A reliable workflow on a cheap server beats a fragile one on premium infrastructure every time.
If you're still learning the build side, start with our step-by-step guide to building an AI agent in n8n, then come back to the hosting decision once you know what the workflow actually needs.
Default to n8n Cloud for client work. It ships faster, the client owns the billing, and you're not the one holding the pager at 2am. Self-host only when a hard constraint forces it: data residency, high volume, or custom code. And when you do self-host, run it on the client's infrastructure and charge for the management. The freelancers who lose money on n8n are the ones self-hosting on a personal server to save the client $20 a month, then eating years of unpaid ops work. Don't be that person.
Want to find the clients who need this work? DevSnipe aggregates AI Agent Development jobs across 5 platforms, with 425 roles posted in the last 30 days alone. Browse the current openings on the AI agent development jobs board and filter for the n8n builds that fit how you want to host.
The software is free and source-available under n8n's fair-code Sustainable Use License (the Community Edition has no execution limit), but running it isn't. You pay for a server, and a production-grade setup with PostgreSQL and backups runs about $24-30/mo on a provider like Hetzner. The bigger cost is your time maintaining it. Note that fair-code lets you run it for a client's internal use but not resell it as a hosted product.
You can, but you shouldn't for production. If it runs on your personal server, you own the uptime forever and you're doing free ops work. Self-host on the client's own infrastructure and charge a management fee, or put them on n8n Cloud so they own the account.
One workflow run counts as one execution regardless of node count, but agents that loop through tool calls still run frequently. A chatbot handling 100 conversations a day is roughly 3,000 executions a month, which already exceeds the $24 Starter plan's 2,500 cap. Estimate volume before you pick a tier.
Self-hosted, every time. Clients in healthcare, legal, or finance often can't route data through a third-party host. Running n8n on their own cloud account keeps the data in their region and under their compliance rules, which is exactly what those contracts require.
A default install on SQLite with no backups is fragile. A proper production setup with PostgreSQL, automated backups, and error handling matches Cloud reliability closely. The gap is about configuration, not the hosting model itself.
Yes. Charge a setup fee for self-hosted configuration (server, database, backups, SSL) and a monthly management fee on top. For Cloud, have the client pay n8n directly and offer a small maintenance retainer. Never fold hosting labor into a build price for free.
A practical, code-first comparison of Make.com and Zapier for building AI agents on client projects. Pricing math, AI feature depth, and which tool you should actually reach for.
A practical breakdown of Zapier vs n8n for AI agent development client work in 2026. Pricing, self-hosting, code access, and which one the paying jobs actually ask for.
A practical comparison of Make.com and n8n for freelance automation work, covering pricing, billing strategies, project values, and which platform builds more profitable long-term client relationships.
Get matched to jobs across Upwork, Skool, and more. Free tier includes daily alerts with 3 full job details per day.
