For months, the honest answer to ‘where does your data live’ was ‘somewhere in the US, I think’. This week I changed the answer.
To explain for new folks reading this: I run a one-woman consultancy, Mission Relearn, and I am building a product, Memnar, in public. AI does a lot of the work behind both, and each week I show you how that actually looks, the wins and the parts that quietly break.
Home in Europe
‘Memnar’, the product I am building, spent its first months on US-based servers, because that was the fastest way to get something real in front of people. The trade-off was that everything, your files, your logins, the emails it sends, sat on infrastructure an ocean away, under someone else’s rules. This week I moved all three of those onto European infrastructure (Mistral, Gladia, Supabase), most of it in a single afternoon.
For a lot of the people I work with, Dutch schools, government teams, European companies, ‘where does the data actually live’ is a question that becomes more and more important. If Memnar holds the way you write and think, kept in your own files, that question matters more here than almost anywhere. Now the answer is short. It is here, in Europe. I didn’t do it to tick a compliance box, it is where I want this thing to live.
The move was not very clean. My email verification kept failing for hours, and the cause turned out to be an old invisible setting still quietly pointing the domain back at the previous platform, hiding all my new records underneath it. One command finally showed me what was really going on.
Note to self, and to you: when something refuses to verify, suspect the thing you set up months ago and forgot.
I stopped trusting my own eyes
Twice this week I caught myself about to argue with a screen. The Memnar homepage looked off on my phone, too much empty space between things, but ‘looks off’ is a feeling, and feelings are kind of hard to fix. So instead of eyeballing it, I let my AI system measure the actual gaps between sections and flag every empty band taller than 200 pixels. Twelve of them. After the fix, zero, with the desktop version left untouched.
So, my tip: decide what ‘good’ looks like as a number before you start, and you save yourself the argument with the screen later.
What I read this week
GLM 5.2 beats Claude in our cyber benchmarks: Semgrep put an open-weight model against Claude on security tasks and it won, at a sixth of the cost. A real data point if you care where your models come from, which after this week I do.
Iwo Szapar on Claude’s Fable 5 skills: six portable working habits pulled out of one model before it goes paid, worth grabbing if you work in Claude Code.
Bahadir Yilmaz’s three daily AI agents: his favourite one hunts for loops that have run too long without a conclusion, a smarter kind of automation than the busywork most agents do.
Margot Lee on journaling and neuroplasticity: a set of prompts built off the neuroplasticity conversation, a calm counterweight to a week of moving servers.
That is the week. A product that now lives in Europe, and a founder who trusts a ruler more than her own tired eyes.
If someone asked you today where your tools keep your data, would you know the answer? If not, that is worth an afternoon. And if you have a ‘looks off but I cannot say why’ problem of your own, hit reply and tell me what it is. I might have a number for it.