ServicesPrivate AI and data residency
Run AI where your data is allowed to be
If your data can't leave a country, a cloud account or your own servers, the AI that reads it can't either. We build and deploy AI inside your environment or in the region you choose, so prompts, documents and outputs stay where your contracts and regulators expect them.
- Your cloud account or your servers
- Tested on your own tasks first
- No training on your data
Four ways to keep AI in bounds
Most clients use a mix: sensitive work on a private model, the rest on a managed model in the right region.
Self-hosted open-weight models
Models such as Llama, Mistral or Qwen running on your own hardware or in your private cloud, served with tools like vLLM. Nothing leaves your network.
Managed models pinned to a region
Enterprise endpoints from Azure OpenAI, Amazon Bedrock or Google Vertex AI, set to process in the regions you approve, with retention and training settings checked against the provider's current terms.
Routing by data category
One gateway decides per request: health data or legal privilege goes to the private model, routine drafting can use the managed one.
Fully on-premises
For the most sensitive work, a deployment with no outside connection at all, including updates delivered through your own change process.
Where the boundary sits
your users
inside your region
outside the boundary
GDPR Art. 44
The boundary can be a country, a cloud region, an account or your own building. The design is the same; only the edge moves.
What keeps it private
Where the model runs is only part of it.
Private networking
No public endpoints. Traffic stays on private links inside your cloud.
Your encryption keys
Data at rest encrypted with keys you manage, so access can be revoked from your side.
Contract terms that match
Data processing agreements, standard contractual clauses for transfers, and business associate agreements where HIPAA applies.
No training on your data
Checked in each provider's terms and settings, and true by design for self-hosted models.
The trade-offs
Private AI isn't always the right call. We test before you decide.
Quality
Smaller self-hosted models can fall short of the largest hosted ones on hard reasoning. On extraction, classification and summarising, the gap is often small. We measure it on your own examples.
Cost
GPUs cost money whether they're busy or not. For low volumes a regional managed endpoint is usually cheaper. We model both.
Upkeep
Self-hosted models need updating and monitoring. We hand over runbooks and can run it for you.
Residency isn't everything
Keeping data in the EU doesn't settle legal basis, purpose or retention. Those still need answers.
Questions
Can we use GPT or Claude models and still meet residency rules?
Often, yes, through enterprise endpoints that process in specific regions. The options change, so we check each provider's current regions and terms against your requirements before recommending one.
Which open-weight models do you use?
Whichever does your task best within your hardware and licence constraints. We compare a few on your own examples and show you the results before choosing.
Who runs it after launch?
Your team, us, or both. You get runbooks, monitoring and the full configuration either way.
Is data residency enough for GDPR?
No. It helps with transfer rules, but you still need a legal basis, a defined purpose, data minimisation and retention limits. We cover those in the same project.
How our live systems handle data
Every case study has a section on how its data is handled: what the AI sees, what it never sees and which rules that maps to.