Memorare Press · First Edition · 2026
Sovereign AI Engineering
How to Own the Agent Stack, from the Model to the Browser
by Miguel Amigot
Every AI quote has two problems, the price and where the data goes, and both are one decision made by someone else about who owns the stack. Sovereign AI engineering is the discipline of making that decision yourself: the code, the data in place, the model as a setting, one identity across every client, and agents in a harness you can inspect. Written for the CIO: twenty chapters and seven appendices on models and the hardware to run them, connectors built on MCP with a broker that holds the credentials, clients that share one sign-in without holding a secret, harnesses and headless applications, and the operating model from the first pilot to the exit you never use.
The organizations that get the most out of AI are not the ones with the largest budgets or the earliest pilots. They are the ones whose leaders understood, early, that the model is a commodity, that the data must stay where it lives, and that the applications their people use should be built on an identity and agent layer the institution controls.
20
Chapters
7
Appendices
30k
Words
~2 hrs
Read time
Table of Contents
The full text of Sovereign AI Engineering is published here, chapter by chapter. Every section is a page of its own — readable, linkable, and shareable.
Preface
Why a CIO Needs an Engineering Book
The Bill Comes Due
Chapter 1
Two Problems with Every AI Quote
The per-seat arithmetic · Where the data goes · Why the two problems are one · What usage pricing changes · The question to ask every vendor
5 min read
Chapter 2
The SaaSpocalypse
What the debate is actually about · What a subscription was paying for · The three things that remain scarce · The headless conclusion · What it means for a CIO's portfolio
5 min read
Chapter 3
What Sovereign Means in Engineering Terms
Four questions you can verify · The deployment tiers · One codebase, four tiers · What sovereignty is not · A worked example
5 min read
Chapter 4
The Lock-In You Did Not Sign
Model lock-in · Identity lock-in · Connector lock-in · Format and memory lock-in · Writing the exit into the architecture
5 min read
The Model Layer
Chapter 5
Models Below the Frontier Are Commodities
What the frontier is for · The six-month rule · Pricing power and where it goes · The design consequence · How to test the claim in your own house
5 min read
Chapter 6
Routing, Not Choosing
Four dimensions of a routing decision · Routing by data classification · Fallbacks and outages · Open-weight models in the mix · What the administrator sees
5 min read
Chapter 7
Hardware You Can Put Your Hand On
Why memory, not FLOPS, is the number · The department tier: Apple silicon · The team tier: DGX Spark and DGX Station · The institution tier: DGX in the data centre · Routers across machines · What to budget and what not to believe
6 min read
Chapter 8
The Air-Gapped Deployment
Who needs it · What has to be inside the perimeter · Updates by signed bundle · What you give up · Why the same codebase must run both ways
5 min read
The Data Layer
Chapter 9
The Migration That Never Ends
How the migration starts · What it actually costs · Why the copy is always wrong · The case for reading in place · What changes when the data stays put
5 min read
Chapter 10
Connectors Built on MCP
What a connector is · Why a protocol and not an integration · The catalogue a CIO should expect · Scoping to the person · Plain-English skills on top of connectors · What to ask a vendor about their connectors
5 min read
Chapter 11
The MCP Broker
The piece most vendors skip · What the broker holds · Deny by default · Scoping every call to the person · The audit line · Why the broker belongs to the institution
5 min read
Chapter 12
The Ontology over Your Systems
The vocabulary problem · A pass-through, not a warehouse · How identity and authorization flow through it · What belongs in the ontology and what does not · Building it incrementally
5 min read
The Client Layer
Chapter 13
One Backend, Every Surface
Where the work actually happens · Seven surfaces, one codebase · What a native shell adds · Shared components, not shared screenshots · The cost of the alternative
5 min read
Chapter 14
Identity Without Secrets in the App
Two kinds of credential · The hosted sign-in hand-off · Why the organization's key never reaches the client · Organizations, switching and visiting · Native shells and the system browser · The threat model, honestly
5 min read
Chapter 15
The Agent in the Browser
What the side panel is · Reading the page · Acting on the page · The refusals and the approvals · The trust boundary · What the demo proves
5 min read
Chapter 16
Shared Components as Policy
Where governance usually fails · The components that carry policy · Building an application in an afternoon · Skills that make a coding agent a platform developer · What the CIO gets
5 min read
The Agent Layer
Chapter 17
What a Harness Is
The loop · Nine attributes of a flexible harness · Memory in three scopes · The sandbox and the network boundary · The same harness on every surface · What to demand
5 min read
Chapter 18
The Five Layers of a Production Agent
The five layers · Why most products ship only the first · Orchestration and memory are not a longer prompt · Governance belongs in infrastructure · The questions to ask about each layer
5 min read
Chapter 19
Headless Applications
The post-SaaSpocalypse application · Managing a platform from a terminal · Skills instead of screens · Where a head is still needed · What it does to the budget and the org chart
4 min read
The practical toolkit
Framework
Framework: the five properties of a sovereign AI stack
Appendix A
Sovereignty Maturity Model
3 min read
Appendix B
The CIO's Readiness Assessment
4 min read
Appendix C
Reference Architecture
3 min read
Appendix D
Hardware Sizing Tables
3 min read
Appendix E
Connector and Broker Policy Catalogue
3 min read
Appendix F
A 90-Day Sovereign Pilot Plan
4 min read
Appendix G
Glossary of Sovereign AI Engineering Terms
5 min read
Run the stack this book describes
ibl.ai is the agent operating system you run yourself: you own all the code and the data, in your cloud, on-premise or air-gapped, with any model, one identity across web, browser, desktop, mobile and terminal, and no per-seat pricing.
Last updated