BxM Consulting
Book a call

In short

We built an environment for automation and AI that runs entirely in Europe: servers in Frankfurt, the language model in the EU, no access key on the server. The point is not the one workflow we built. The point is that every further one slots in without setting up anything new. Build the foundation once, then add on. We build the same environment in your AWS account. You pay for usage instead of licenses. If you withdraw our access, everything keeps running.

The setup as a whole: an AWS account in Frankfurt, an EC2 instance running n8n as the automation engine, and the language model reached through a role instead of a key. Simplified: network and security groups are left out for readability.
Account on Amazon Web Services in Frankfurt with one server running n8n and three more containers, connected on the left to the company's systems and on the right, through a role, to the language model

The setup as a whole: an AWS account in Frankfurt, an EC2 instance running n8n as the automation engine, and the language model reached through a role instead of a key. Simplified: network and security groups are left out for readability.

What it compares to

Ready-made platformYour own environment
Your data sitswith the vendorin your account, on your contract
You paya license per user, modules one by onefor actual usage
One more employeecosts extracosts nothing
One more workflownew module, new pricegets added on
If you stopaccess endskeeps running, without us

What runs today

  • A confirmation email for every contact request, since July 31, with nobody watching.

  • Error alert: if something fails, an email arrives within seconds with the workflow, the error and a link to the run.

  • Meeting minutes from a transcript, as a PDF in house style. Built and tested. We switch the chain on once the impact assessment, contracts and consents are in place. In that order, and the same goes for you.

The finished minutes from a synthetic test transcript.
First page of meeting minutes with a summary, decisions, a table of tasks and deadlines, and the open points

The finished minutes from a synthetic test transcript.

What you can connect

AreaExamples
Email and calendarGoogle Workspace, Microsoft 365, Exchange
FilesGoogle Drive, SharePoint, OneDrive, Dropbox, network drive
Customers and salesHubSpot, Pipedrive, Salesforce, Monday, Zoho
Tasks and projectsAsana, Jira, Trello, ClickUp, Notion
Accounting and ERPDATEV, BMD, sevDesk, lexoffice, Odoo, SAP Business One
Inventory and shopShopify, WooCommerce, Shopware, JTL
CommunicationSlack, Teams, WhatsApp Business
SupportZendesk, Freshdesk, Jira Service Management
HR and timePersonio, BambooHR, Clockodo
DatabasesPostgres, MySQL, MongoDB, MS SQL Server
Signatures and formsDocuSign, Adobe Sign, Typeform

Examples of systems with an open interface, not ready-made integrations. If your software has one, connecting it takes hours, not weeks.

Connected on our side: a file storage and an email service.

  • Quotes from requests: key data captured, draft in house style, a person signs off.
  • Create and match invoices, with consecutive numbering.
  • Sort the inbox: request, complaint or advertising. The AI suggests, a person decides.
  • File documents: a letter arrives, gets read, named and filed.
  • Minutes and tasks: minutes from the conversation, tasks from the minutes.
  • Make knowledge searchable: quotes, contracts, manuals. In plain language, with the source named.

Each of these is an add-on to the same foundation. No new module, no second rollout project. What is worth doing is not something this list can tell you. A look at your workflows can.

What it costs

You pay for what runs. The infrastructure (server, storage, fixed address) costs around 40 euros a month before a single workflow has run. Everything else is model calls, billed by usage.

An additional employee costs nothing extra. There is no license per user and no feature that is unlocked separately.

How it is built

One AWS account in Frankfurt, one small server, four containers: a database, the automation engine, a service that renders documents in house style, and a web server. Everything exists as code. The same code builds the environment a second time for you.

The server holds no key for the AI. It may assume a role that can call the language model. The permission comes fresh from AWS with every call and expires after minutes. That role knows EU regions only. A call outside the EU does not fail because of an agreement, but because of a missing permission.

No SSH access, no key pair, an encrypted disk. Maintenance runs through AWS and is checked anew on every access.

The whole chain. Both branches end in the same quarantine folder.
Chained workflow steps on a canvas, from the schedule trigger on the left to the file storage on the right, with two branches leading to a shared end point

The whole chain. Both branches end in the same quarantine folder.

Why it runs reliably

A workflow is not a request to a model but a chain of steps you can check one by one. Two of them are gates:

If the text is too short, the language model is not even called. The file moves to quarantine.

If the layout service reports no success, the same happens. Nothing lingers, nothing is guessed.

We made both fail on purpose. Layout service switched off in live operation: no document, source file untouched. Confirmation email broken on purpose: the alert arrived within seconds.

Successful runs are not stored at all, otherwise transcripts would sit in the log as plain text. Errors are deleted after 72 hours.

Where AI works, and where not

The AI writes text. Nothing else. Finding, checking, filing and sending files is all fixed logic. No model decides whether an email goes out.

If a connected service fails in the middle of a run, the run stops. That is intended: nothing is half processed, the files stay untouched and are picked up again in the next cycle.

How we set this up for you

You create a role for us with named permissions. In your console you see what it may do, and you delete it there with one click. If you like, we keep looking after the environment through the same role, not through a second access. Account, server and workflows are in your name and stay there.

We show you the running environment in a call.

“Your data in your account” is a structure. “We’ll take good care” is a promise.

Ask about a similar project