Private AI questions, answered
Plain answers to the questions regulated organisations ask most about private and on premises AI.
Last updated 15 October 2026
Book an assessmentAbout private AI
What is the difference between private AI and a public AI service?
A public AI service receives your documents, processes them on its own model and infrastructure, and sends answers back. A private AI platform keeps every step, from documents to answers, inside the cloud or data centre you already control and audit.
Is ChatGPT safe for confidential data?
It depends on the version, the terms you have agreed and the data. Consumer tools may keep and use what you type under the provider's terms, while business tiers add contractual controls. In every case your documents are processed on another company's service, outside your own audit and change process. Our guide, is ChatGPT safe for confidential data, covers this in detail.
What is sovereign AI?
Sovereign AI means you control the whole AI system: where the data is stored, where the model runs, who can access it and who can change it. Data residency is only part of that. Read what sovereign AI means for the full explanation.
What is RAG, and why does it matter?
Retrieval augmented generation (RAG) finds the relevant passages in your own documents and gives them to the model with the question, so answers draw on your material. In a private platform, the documents, the vector database and the model all sit inside your estate.
We already have Microsoft 365 Copilot. Do we need this?
Perhaps not. Microsoft 365 Copilot suits everyday work in Microsoft 365: it answers from the email, files and chats each person can already open, and Microsoft states that prompts and responses are not used to train its foundation models. A private platform is worth considering when you need to choose the model yourself, search data held outside Microsoft 365, give collaborators their own scoped collections, decide what is logged and where, or handle work your organisation has not approved for Copilot. Many organisations use both. See where the two differ.
Is a private LLM as good as ChatGPT?
For many work tasks, such as answering questions over your own documents, summarising and drafting, current open weight models perform well, especially with retrieval over your documents. It varies by task, which is why a proof of concept measures the result on your own data before you scale.
Data, privacy and UK GDPR
Will our data be used to train a model?
Not if the platform is scoped to prevent it. With a model on compute you control, your documents are used only to answer your questions inside your own environment, and any fine tuning you later choose happens inside your estate. With a managed model, check the provider's terms; Microsoft, for example, states that Azure OpenAI prompts and outputs are not used to train its models.
Does a private AI platform make us GDPR compliant?
No platform makes an organisation compliant on its own. A private platform keeps personal data inside the environment you already control, and we document the architecture so your data protection and security teams can use it in a DPIA and in their own assurance.
Can we use Azure OpenAI and still keep data private?
Yes. An Azure OpenAI model can be deployed in your own Azure subscription, as a regional pay as you go or provisioned deployment in a UK data centre, and reached only over private networking. Prompts and answers are then processed in that UK region, and your data stays in your tenancy, under your agreement with Microsoft. Global and Data Zone deployments can process prompts in other regions, such as the EU or the US, and often reach the newest models first; we use them only if your security team chooses to allow it. Abuse monitoring also needs a decision. Where your requirements rule out a managed model, the model can run on compute you control instead.
What can our security team review before we commit?
Our assurance page sets out how a platform handles data, identity, logging, change and handover, with an illustrative architecture and the headings of the architecture pack your reviewers would receive.
Which standards do you design to?
We design to the CIS Benchmarks and Critical Security Controls, the NCSC Cloud Security Principles and the Microsoft cloud security benchmark for Azure workloads, and access follows Zero Trust principles. We document the design so your team can map it to the frameworks you report against, such as ISO 27001 or Cyber Essentials.
Running a private platform
Can you run a large language model on premises?
Yes. Open weight language models can run on GPU servers in your own data centre or on GPU machines in your own cloud tenancy. Documents, embeddings, the model and its answers can all stay inside your boundary. Our guide to running an LLM on premises explains what it takes.
Do we need to buy GPUs?
Not necessarily. A private platform can run on GPUs in your own data centre, on GPU capacity in your existing cloud tenancy, or on a hybrid of both. The right choice depends on your workload, your security requirements and the estate you already have.
How much does a private LLM cost?
It depends mainly on the size of the model, how many people use it, whether it runs on hardware you own or on cloud GPUs, and how much of the running you want us to do. An AI readiness assessment sets out these choices for your situation before you commit to a build.
Who owns what you build?
You do. The code, configuration and documentation we write are yours. We supply no applications, data or services of our own: we use only services you provide, or that you allow us to buy on your behalf, and every licence, account and detail is in your organisation's name. Infrastructure and pipelines are defined in code and deployed through your own change process, and we operate the platform alongside your engineers or hand it over fully documented.
Who will we work with?
You work directly with the senior architect who designs your platform. Any other specialist is named in the scope and joins only with your agreement.
Working with us
Where should we start?
Most organisations start with an AI readiness assessment. It reviews your data, infrastructure and security position, ranks AI use cases by value and by risk, and ends with a prioritised plan. To prepare, you can answer six scoping questions and bring the brief.
What happens after we get in touch?
A first discussion, then a written scope, then agreed decisions on access and data handling before any of your data is read. How an engagement starts sets out each step.
We already have a prototype. Do we have to start again?
Usually not. We keep what works, such as the prompts, the retrieval approach and the evaluation questions, and rebuild what it runs on: hosting, identity, data connections and operations, inside your own estate. The prototype to operated platform map lists the ten decisions involved.
Do you only work in Oxford?
No. We are based in Oxford and work with organisations across Oxfordshire, the Thames Valley and the rest of the UK.
Still have a question?
Ask it here, with a little about your data and your estate, and we will come back with an answer for your situation.
Thank you. Your request is on its way. We will reply from info@tlgsolutions.co.uk to arrange a first discussion.
That did not send. Please email info@tlgsolutions.co.uk instead, and we will reply from there.