Use AI. Keep control of confidential data. Private AI on your servers or with our hosting partners.
You want to work with internal documents without uploading them to arbitrary AI services. We establish what the system needs to do, select appropriate infrastructure and set up the workspace: on-site, on dedicated servers or with carefully controlled cloud access.
Shared rules for confidential data and AI
One person summarises a customer email in a personal account; another uploads a document to try a new tool. Without a shared workspace, it is hard to know what is processed where.
We start with the tasks and data. Some can use an approved business AI service; confidential documents may need a private environment with defined permissions and rules for external connections.
Three ways to run private AI
On your own hardware
Documents are processed inside your environment. A Mac Studio may suit limited tasks; larger models or concurrent use may need an NVIDIA GPU server. We assess the need before any purchase.
Dedicated hosting
A separate server environment with ISO 27001-certified partners, and no hardware in your office. Location, access, responsibilities and costs are agreed in the proposal.
Hybrid, with controlled release
Protected processing stays private; a powerful cloud model can assist with selected tasks using approved content. The boundary is configured deliberately.
Choose the task before the hardware
-
01
Define tasks and data
Search, draft, classify or automate a step? We identify source documents, confidentiality needs and who checks the results.
-
02
Size the system
Model requirements, document volume, response time and simultaneous users affect capacity. We compare local hardware with dedicated hosting, including operating costs.
-
03
Configure and test
We connect the agreed sources, set permissions and test real tasks with your team, including training on checking answers.
-
04
Agree ongoing operation
Updates, maintenance, backups and support can be added, with responsibilities assigned between you, us and any hosting partner.
What a private environment can support
Find internal knowledge
Ask about approved manuals, policies or project material, with sources and the agreed access rules.
Prepare documents
Summarise long material, locate passages and prepare drafts for your team to check.
Automate repeat work
Smaller local models may be enough to classify documents or prepare responses, with reviews and a route for exceptions.
Use cloud AI selectively
Complex work without confidential inputs can still use a cloud model, kept separate from internal-only tasks.
Privacy depends on the whole setup
Running a model locally does not by itself settle privacy and confidentiality requirements.
Access rights, logs, retention, external connections, maintenance access and your own requirements also shape the design.
Removing names does not always make a document anonymous. We agree what may leave the private environment; internal-only tasks are configured without a cloud model. A hosting partner’s ISO 27001 certification covers its certified scope, not your specific application.
An example: private customer enquiry, general cloud template
-
Keep the confidential enquiry internal
A local model reads the approved message and finds internal material; customer names, contract details and attachments stay inside.
-
Give the cloud a separate, approved task
A cloud model might draft a general checklist for the response, without the original message or customer details, as a defined handover.
-
Review the actual response
The template is completed internally, and a responsible person checks content and recipient. If the task needs confidential context, this step stays internal too.
Questions about private AI
What does on-premise mean?
The AI system runs on infrastructure you control and can process documents without sending them to an external provider. We clarify whether updates or support need connections.
Do we need a large server?
Not necessarily. A focused application for a few people needs different capacity from a system serving many simultaneous requests.
What does private AI cost?
Simple internal chatbots can start at €2,000; on-premise deployment may add hardware and operational setup, and hosting brings recurring costs. The proposal separates development, infrastructure and support.
Can we still use ChatGPT, Copilot or Claude?
Yes. A suitable business version can remain available for approved work. We train your team on which information may be used with each tool.
Can you maintain the system?
Yes, we or our partners, depending on scope. Responsibilities and costs are agreed before work begins.
Which information needs to stay private?
Describe your tasks and requirements in a free first call. We will help identify a suitable starting point.