n8n consulting and development

1. 01

   You want to roll out n8n and need someone to set up hosting, access rights and structure properly from the start.

2. 02

   Workflows already exist, built by someone who has left or has no time. They run, but nobody knows why.

3. 03

   A workflow needs more than the built-in nodes offer: custom code, a backend or a language model with a review step.

4. 04

   Your IT demands traceability before a workflow goes live: versioning, environments, monitoring.

1. 01Intro call

   Map the process

   We walk through the process as it runs today: input, steps, exceptions, who approves. The result is a written effort estimate.

2. 02Foundation

   Instance and structure

   Hosting, access rights, test and production environments, versioning of the workflows. For existing instances, an inventory and clean-up come first.

3. 03Build

   Workflow with checkpoints

   Implementation with error handling, idempotency and alerts. Where a language model classifies or drafts, the flow stops at the sensitive points for approval.

4. 04Handover

   Documentation and walkthrough

   Every workflow comes with the documentation your team needs to adjust it, and a walkthrough. Ongoing support by effort if you want it.

The outcome is workflows treated like software: versioned in Git, tested in a staging environment, with alerts that report before a customer does. Your team can adjust them without asking me each time. Whatever does not belong in n8n, because it runs too long or computes too much, moves into a small backend that the workflow calls.

What does n8n consulting cost?

Per project against an effort estimate, which you receive in writing before commissioning. A single workflow takes days; rolling out an instance with several workflows takes weeks. I quote the day rate in the intro call.

Self-hosted or n8n Cloud?

Self-hosted when data has to stay in-house, custom binaries are needed or your IT runs the instance anyway. n8n Cloud or a managed instance when nobody wants to look after a server and the data allows it. Where the managed variant reaches its limit I describe on the blog, using video processing as the example.

Can you take over existing workflows?

Yes. The first step is an inventory: which workflows run, which have failed silently, where credentials sit in plain text. Then we decide what gets cleaned up, what gets rebuilt and what gets switched off.

Where do language models come in?

Where a rule is not enough: classifying free text, pulling data from unstructured documents, drafting replies. Each of these steps gets a checkpoint, and whatever the model cannot classify with confidence goes to a human.

Proposal after a free intro call