Part of the Sovereign Stack · Family: Sovereign

Private AI

AI models run on your company’s server and are available only to authorised employees. Search internal documents, draft text or transcribe meetings in your company’s private environment.

AI reserved for your company

Use the models through accounts reserved for authorised employees.
Keep documents, questions and answers in your company’s environment.
Your company controls who can access the models and information.

How it works

Choose useful tasks and documents, then check the permissions and hardware.
Run the selected model and document-search software on the local server.
Give authorised employees private access to the models and internal documents.

Scroll sideways to read the full drawing.

The steps above explain this drawing. Terms are described below.

Private AI models for your company.

Your people and documents: Choose useful everyday tasks
Select the right documents
Agree who may use them
Private AI: Models run on your server
Search your own documents
Check licences and versions
AI for your company only: Authorised employees only
Models and data stay internal
Your company controls access

Your company decides who can use the models and access its information.

Open weight: Model files available to download under their stated licence.
Open source: Software whose licence permits the specified use, inspection, modification and sharing.
API: Application Programming Interface. A connection through which your applications communicate with the local model server.

Documents, access and privacy

Private AI is an internal service for your company. Only employees you authorise can use the models. Your company controls access to the models, documents and conversations.

We decide with you which documents the system may use and which people may access them. The local setup is designed to process the agreed documents and questions on your server.

Backups, maintenance, software downloads and any optional external connection are reviewed as part of the deployment. Local processing needs clear access controls as well as a local machine.

What to expect from the answers

AI can produce inaccurate or incomplete answers. Check important statements against the original document, and have a person review drafts before they are sent or used for a decision.

We try representative tasks with your team to choose a model and decide where it is useful.

Software and model licences

Open-source software comes with a licence that defines how you may use, inspect, modify and share the program.

Open-weight models publish the trained model files for download. Their licences can still impose restrictions; publishing the weights does not necessarily make the whole model open source.

We check the licence and permitted use of each selected component before deployment.

Choose a useful first task

Start with a bounded task, such as searching an internal manual or drafting a meeting summary. We check document quality, hardware, permissions and the standard of answer your team needs before expanding the use.

Technical details

For IT people: on-premises large language models, open-source serving stack, retrieval over your own documents.

Training for your team

Length: One day, hands-on
For whom: Management, staff and IT teams
On the programme: Test a useful local task, review answers and prepare internal AI rules.

Workshops →