BRING AI TO YOUR DATA.
DON'T SEND YOUR DATA TO AI.
We convert suitable GPU-equipped computers into ready-to-use private AI systems. Chat, analyse documents, write and code with AI running on your premises — and, when required, isolated from the internet.
Move the AI. Not the sensitive data.
Instead of routinely copying confidential material into a public AI service, run suitable models beside the information you need them to work with.
For teams where data leaving the premises isn't acceptable.
Private AI is particularly useful where confidentiality, intellectual property or internal information makes public AI services difficult to use.
Solicitors & legal teams
Discuss contracts, correspondence and case documents with local AI while keeping sensitive client material within your controlled environment.
Accountants & finance
Summarise reports, analyse documents and draft correspondence without routinely uploading client financial information to a public AI service.
Developers & engineering
Let AI explain, generate and review code locally. Proprietary source code and technical IP do not need to be sent to an online coding assistant.
HR teams
Work with policies, CVs and internal documents using AI running inside your own environment.
Document-heavy businesses
Ask questions of manuals, policies, contracts and reports. We can add private RAG/document knowledge capability where required.
Security-conscious teams
For organisations that want the option to disconnect the AI environment from the internet after installation and model provisioning.
We install models for the jobs you actually need.
The exact bundle is matched to your hardware, use case and applicable model licences. Examples include:
Best for: reasoning, analysis, business writing and technical work. OpenAI's open-weight GPT models can run on infrastructure you control; we select 20B or 120B according to the hardware.
Best for: reasoning, maths and coding. DeepSeek models can be downloaded and run locally, so using the local model does not require sending prompts to DeepSeek's hosted service.
Best for: general AI, reasoning, multilingual work and coding. We choose the appropriate Qwen model and size for the customer's hardware and policy.
Best for: efficient everyday AI and software development. Mistral offers open-weight models including specialist coding options such as Devstral.
Best for: general business assistance, drafting, summarisation and document work. A widely used open-weight model family with several sizes available.
Your hardware. Your data. Your choice of AI. Model origin is shown clearly and customers can exclude model families their organisation does not permit. Model availability, licensing and suitability change over time; final selections are agreed during setup. Embedding models for private document retrieval/RAG are installed where required.
We turn suitable GPU hardware into a usable AI computer.
This is installation, configuration and handover — not simply downloading an app.
Need more than the standard build?
We can add components where the requirement justifies them rather than burden every installation with unnecessary software.
- Qdrant/vector database for larger persistent knowledge collections
- RAG/document ingestion for private document Q&A and knowledge retrieval
- Multiple users and local-network browser access
- Specialist models for coding, reasoning or particular workloads
- Ongoing support package available on request
Not sure what hardware to buy?
Contact us first. Tell us your use case, number of users and budget and we'll advise on an appropriate GPU system.
Use your GPU computer, or buy one ready to go.
One-off private AI installation and configuration.
- Core local AI software stack
- Models selected for your requirements
- Document/RAG capability where required
- Offline/air-gap configuration where required
- Full documentation
- 4 weeks email support
ASUS Ascent GX10 supplied with our private AI environment installed, configured and tested.
- ASUS Ascent GX10 • 128GB unified memory
- Complete local AI software setup
- Curated model bundle
- Private document/RAG capability as agreed
- Full documentation
- 4 weeks email support
What does “air-gapped AI” mean here?
Can the system really run without the internet?
Yes, provided the selected models and required dependencies have first been installed locally. Where required, we prepare the system for offline operation. True air-gapping also depends on how your organisation configures and isolates its network.
Can you install on something other than a GX10?
Yes. The £750 service is for suitable GPU-equipped PCs and workstations. Contact us before buying hardware and we can advise based on the models, workload and number of users you expect.
Do I automatically need Qdrant?
No. A vector database is useful for some larger or persistent RAG/knowledge workloads, but it is not required for every local AI installation. We install it when the use case benefits from it.
Can we choose which AI models are installed?
Yes. We explain the origin and typical use of the available models and agree the bundle with you. If your organisation does not permit a particular model family or country of origin, we leave it out.
Can it help our developers code?
Yes. We can install suitable coding models so source code can be analysed and generated locally rather than routinely sent to an external AI service.
Is it automatically GDPR or professionally compliant?
No technology purchase alone creates compliance. Local processing can give your organisation greater control over where data is processed, but your own policies, security, legal and professional obligations still apply.
What happens after installation?
You receive full handover documentation and four weeks of email support. Ongoing support packages are also available on request.
Bring AI to your business.
Already have GPU hardware? We'll tell you whether it is suitable. Starting from scratch? Tell us what you want the AI to do and we'll help you choose the right system.
hello@digitalpoint.co.uk