The build.
The Sovereign AI build
Once the data's in shape, this is where the tools get made: the private AI assistants your staff use day to day, built and deployed inside your own infrastructure. It feels like the public tools they're already used to, but nothing leaves your control.
The Sovereign AI capabilities on this site show what a build can do. This page is about how one actually comes together, what you end up owning, and what it costs to run.
What gets built
What your staff see is a private AI assistant: a chat tool that looks and feels like ChatGPT, with document upload, a separate workspace for each job, and answers that cite their source. The difference is where it runs and what it's allowed to do.
Underneath, it's an open weights model running on AWS Bedrock in London, wired to your own documents through a retrieval layer so it answers from your material rather than guessing. Each use case is a separate workspace on the same build, so one assistant can cover several jobs at once.
The three layers, in practice
A build is sovereign on three fronts at once. This is what each one means once it's actually running:
- Data sovereignty. Your prompts, your documents, and the answers all stay in your own AWS account, in your tenancy, in the UK. Nothing is sent to a US AI company.
- Model sovereignty. The model is open weights, something you can see inside and run yourself, not a closed service you can only call across the internet.
- Operational sovereignty. You can audit it, change it, and switch it off. The whole thing is described in code you hold, so you're never dependent on me to keep it running.
Each of these is covered in more depth in What is Sovereign AI?
Starter and production
A build can take one of two shapes. They share the same content, workspaces, and model. Only the hosting underneath changes.
Starter
Everything runs on a single server. It's the simplest and cheapest shape, and the right place to start for a pilot or a small team. Around ten AWS resources: one EC2 instance running Docker, with Caddy handling HTTPS and Route 53 for the web address.
Production
The same assistant, run across more than one server for reliability, with AWS handling patching, scaling, and replacing servers if one fails. Around forty resources: ECS Fargate behind an Application Load Balancer, with shared storage underneath.
You can start on the starter shape and move to production later, without rebuilding from scratch.
How it lands in your infrastructure
The whole build is described in Terraform, so the same configuration that runs in my account runs in yours. The tools appear in your AWS account, identical to what you've seen, and they stay there.
Single sign on connects the assistant to the accounts your staff already use, through Cognito and your existing identity provider. No new password, and access ends automatically when someone leaves. Your documents stay current through a small sync from wherever they already live, so the assistant never goes stale.
What you get at the end
A working private assistant your team can use, running in your own infrastructure. The Terraform that describes it, so you're never locked to me. And a handover, so your own people can run it and change it.
You own all of it. If you ever wanted to move it to a different cloud, or take it fully in house, nothing here stops you. That's rather the point.
Time and price
A first build is usually a few days of data and prompt work, plus a couple of days to set up the infrastructure. Later builds are faster, because they reuse the same foundation. The exact shape depends on the use case and the state of your data, which is what the Discovery and the data work settle first.
Running costs are modest: tens of pounds a month for a starter deploy, more for production, with small per question model costs on top. You see real numbers before you commit to anything.
"Working with Peter was an absolute pleasure, thanks to his flexibility, excellent communication skills and honesty."
Luca Russo, Web & Social Collaboration Functional Lead, Givaudan SA
Where this fits
A build usually comes after a Discovery and, where it's needed, the data work. If you'd rather see what one can do before anything else, the Sovereign AI capabilities page is the place to start.
Or email me at peter@peterbrady.co.uk with a short note about your organisation, roughly how many people are in it, and the sector you're in. I'll come back with a time for a thirty minute scoping call.
