Notes
Plain-language pieces on private and sovereign AI: what leaves your network, what stays, and how to choose.
Which Open Model Should You Use?
There are now dozens of serious open-weight models, and most comparisons rank them on a single number. This guide starts with a plain-language answer to 'which one should I use?', then gets steadily more technical: a full comparison table, rankings on six different axes, the licence small print, and the memory arithmetic that decides what you can actually run.
The Case for Private AI
Companies are not choosing whether to use AI. Their employees already are. The real choice is where that AI runs, and the case for keeping it inside the company has never been stronger.
Open Models Briefing: The October Surge
Between early October and early November 2026, open-weight AI had its busiest stretch on record: a Chinese model briefly within three points of the frontier, three Western releases in the hundreds of billions of parameters, and a licence landscape that matters more than ever.
Apertus, One Year On
Fourteen months after its launch, Apertus has moved from research artefact to infrastructure: in production at a canton, available in mainstream chat apps, and extended to version 1.5. What it is, what it isn't, and why it matters more than its benchmark scores suggest.
Open vs Closed: Reading the Four-Month Gap
Open-weight models now trail the closed frontier by months, not years. But an average hides a lot. On some tasks the gap is gone; on others it is wider than the headline suggests. Here is how to read it.
What Open Weights Actually Buy You
The debate about open AI models is usually conducted in the language of ideology. The business case is more concrete: open weights give you custody of the model, control over when it changes, and economics that improve with scale.
The Case for a Local RAG
Cloud assistants know everything about the world and nothing about your work. A retrieval system running on your own hardware inverts that trade, and for a surprising number of tasks, it is the better one.
What “Sovereign AI” Really Means
'Sovereign AI' is now printed on every vendor brochure in Europe. Most of them mean something narrower than you think. Sovereignty is not a property of a product; it is a set of answers about who controls each layer of the stack.
Every Prompt Is a Data Transfer
Most organisations spent a decade learning where their data lives. Generative AI undid that work in eighteen months, because every prompt is a transfer of data to someone else's computer. Here is how to think about it clearly.
Sounds like a challenge for your company?
Get in touch→