n8n agency: AI workflows
hosted in France or on your own infrastructure.

Not sure yet what to automate, or with which tool? Start with the Consulting practice’s process automation: there we map your tasks and put a number on the gain before choosing the tool. If you want n8n on your own infrastructure, the rest of this page is for you.

Two questions decide it: can your data pass through a service you don’t control, and what volume will you be working at in two years?

If the answer to the first is no, or if volume makes usage-based billing unreasonable, n8n becomes the right tool. It is installed on your own infrastructure, doesn’t bill per operation, and its code is open. In return, it requires available skills that most organizations don’t have in-house, and that is what we bring.

n8n is one of the building blocks of Bonne Idée’s AI practice: it also triggers the AI agents we build.

This is the first question to settle. It is about what you are willing to entrust to others, and about who will look after the machine.

The three reasons that tip the balance toward hosting in France or on your own infrastructure

• The nature of the data. Health records, accounting documents, human resources data, information under contractual constraints. When a clause forbids you from letting the data leave, the decision is already made.
• Volume. Online platforms bill per operation or per task. An eight-step workflow triggered two thousand times a month uses sixteen thousand operations, not two thousand. Past a certain threshold, the subscription becomes comparable to the cost of dedicated infrastructure.
• Reversibility. An n8n workflow belongs to you, can be exported and can be reinstalled elsewhere. On a closed platform, leaving means rebuilding everything.

The three reasons to stay on an online platform

No one in-house to keep an eye on a server. A low volume, where the subscription costs less than an hour of maintenance a month. Or a need for very specific connectors that are better covered elsewhere.

An installation hosted in France or on your own infrastructure, in an organization that has no one to keep it running, creates one more dependency.

What hosting in France or on your own infrastructure really involves

A machine to keep up to date, tested backups, monitoring that alerts a human, and a restore procedure that someone has already run at least once. The workload is modest, and it lasts as long as the instance does.

That is what we take on when you entrust it to us, and what we document when you want to take it back.

Not every repetitive task deserves to be automated. A workflow that saves two hours a month but costs one in maintenance only gives back one hour.

The calculation we do before building

Three numbers, and they are set down before the first line of a workflow. How many times a month the task occurs. How long it takes each time. And what a mistake costs when it goes unnoticed.

The third is the one people forget and the one that decides. A task that is rare but whose mistakes are expensive deserves to be automated before a task that is frequent and inconsequential.

The flows that pay off most, in our experience

• Creating a record from the place where the information really arrives: a form, a shared inbox, a chat channel. As long as it has to be copied over, someone will forget.
• Reconciliation between two tools that don’t talk to each other, typically invoicing and sales tracking.
• Follow-ups, the most hated and worst-done task because it relies on memory.
• Consolidating a figure that several people each recalculate on their own, with different results.

What we refuse

Automating a process that no one has ever written down. An automation freezes the way things are done: building it on a fuzzy process amounts to setting the fuzziness in stone, and making it harder to fix than before.

Automating a decision that calls for human judgment. You can prepare the decision, gather the facts, alert the right person. Making the call in their place produces errors that no one sees going by.

This is where automation projects differ the most, and almost no quote mentions it.

An automation always fails, one day

A connected tool changes its interface, a credential expires, a third-party service is down for ten minutes. That is how a system that depends on other systems works.

So you have to decide in advance what happens when the failure comes. A silent automation that goes down on a Friday evening produces incomplete data all weekend, and no one finds out until the customer reports it.

An alert on every workflow that touches a customer or an invoice

Any automation that touches a customer or an invoice raises an alert that a human sees, in addition to the line written to the log.

In practice: an alert in the tool the person concerned already opens, with the name of the workflow, what it was doing, and what still has to be redone by hand. An alert that only says « error » leaves all the work still to do.

Recovery, which matters as much as the alert

A flow that fails midway leaves an intermediate state. So we design workflows so that a retry doesn’t create a duplicate, and so that what has already been done isn’t done again. It is invisible when everything is going well and decisive the day something breaks.

What we always document

Each workflow has an explicit name, a description of what it does and of what happens if it fails, and the list of tools it touches. Ten minutes at build time, and the difference between an automation that survives someone leaving and one that dies with them.

We deliver all three, which lets us recommend without a commercial agenda. Here is the trade-off as we frame it.

What n8n gains

Control over the data, since everything stays on your own infrastructure. Cost at volume, since it doesn’t bill per operation. Reversibility, since workflows can be exported. And real flexibility for special cases, because it accepts that you write a little code when that is the shortest path.

What it loses

Getting started takes longer. The connector catalog is thinner for mainstream tools. And above all, it adds infrastructure to maintain, which weighs heavily on an organization with no technical skills available.

The tipping point, in practice

As long as the flow reads like a sentence, « when this happens, do that, then that », all three will do and the simplest one wins. As soon as you have to write « unless », « for each item » or « while waiting for », the visual representation stops being a convenience and becomes a necessity.

We have detailed this trade-off, with billing figures to back it up, in our n8n or Make comparison.

This is the trade-off of hosting in France or on your own infrastructure, and it almost never appears in quotes. An n8n instance is a service exposed to the internet that holds the keys to all your connected tools. Having it compromised costs far more than a compromised brochure website.

What we always put in place

• Restricted access. The admin interface isn’t open to the whole world, and authentication is strengthened. It is the measure that costs the least and keeps out most automated scans.
• Segregated credentials. Each connected tool gets a dedicated account with minimum permissions, never an executive’s administrator account. An automation that reads invoices doesn’t need to be able to delete them.
• Secrets stored in the tool’s vault, never written in plain text in a workflow. A workflow gets exported, shared and copied; a credential that appears in it in plain text travels with it.
• Tested backups, covering both the workflows and their execution history. Until a backup has actually been restored, you don’t know whether it works.
• Monitoring that alerts a human when the instance stops responding, not only when a workflow fails.

Updates, which are the real issue over time

An instance hosted in France or on your own infrastructure doesn’t update itself, whereas an online platform takes care of that for you. Security patches get applied, or they don’t. In the second case the instance keeps working normally, which is exactly the trap: nothing flags the delay, until the day a known vulnerability is exploited.

So we apply a fixed schedule rather than handling things as alerts come in, with a test environment before production when the number of workflows justifies it. A critical patch is the exception and is applied without waiting.

What happens if the instance goes down

Missed triggers don’t catch up by themselves: we document what has to be redone by hand.

Who carries all this

If no one is available in-house, we carry it, as a separate line item, priced on the number of workflows in production. If someone is, we train them and leave them the written procedure. What we refuse is the in-between: an instance handed to a team that has neither the time nor the procedure always ends up degrading in silence.

Two automation quotes can only be compared once you know what each one includes. Here is our breakdown. The project itself gets a custom quote, after a first call.

What is always included

• A survey of your flows as they run today.
• The decision between hosting in France or on your own infrastructure and a platform, in writing, with its consequences.
• The workflows, tested, with their error handling and alerts.
• Documentation in French, and training for the person who will take over the instance.

What is quoted separately

Hosting of the instance, in your name, which depends on your volume. Subscriptions to the connected tools, for the same reason. And custom development, when a business tool offers no standard way to connect. Each has its own line on the quote.

Monitoring over time

An instance hosted in France or on your own infrastructure needs someone to watch over it. This monitoring can also be part of the Partner plan, within ongoing support for your automations.

Keeping your data on your own infrastructure is often the very reason for choosing n8n. The rest is written into the contract: a data processing agreement (DPA), hosting on your own infrastructure or in France, no model training on your data when a workflow calls an AI, and a record of what each workflow processes.

What GDPR and the EU AI Act require of workflows that call an AI is covered in detail on our AI guarantees and data sovereignty page.