Private AI Platform

Your own AI.
Inside your own walls.

A ChatGPT-style assistant for your team, running on a server your company owns. It answers from your files and runs your automations.

What you get

Everything ChatGPT does.
Nothing leaves.

Chat workspace

Your team chats with AI, signing in with their Microsoft account.

Answers from your files

Ask about contracts, policies or client files and get answers with sources.

Automations

Your private AI drafts, sorts and summarises inside your workflows.

Your own server

Hardware your company owns, installed and set up by us.

Access control

When someone leaves, their access goes with them.

Monthly care

Monitoring, updates and support for one fixed fee.

Why private

Built inside your own walls.

It runs in your building

The AI model, the chat and the automations all run on hardware you own.

Nothing trains on your data

Your documents never go to a public AI provider.

Your team signs in as usual

Microsoft work accounts, so access stays under your control.

How we keep it secure
YOUR BUILDING Chat workspaceYOUR TEAM, MICROSOFT SIGN-IN Private AI modelON YOUR OWN SERVER AutomationsMAIL, INVOICES, APPROVALS PUBLIC AI Cloud chatbots

At a glance

Runs on
Hardware your company owns
Includes
Private AI, chat, automations
Live in
About 3 months
Price
Fixed setup fee + monthly retainer

For your IT team

Under the hood.

Proven open-source building blocks, set up on hardware you own. No black box and no lock-in: your IT team can see and check every part.

  1. 01 · Access

    Microsoft sign-in

    Single sign-on with Microsoft Entra ID over OpenID Connect. Your two-step login and Conditional Access rules apply exactly as they do today.

  2. 02 · Workspace

    Chat interface

    A proven open-source chat workspace such as LibreChat, served over HTTPS on your own network. Conversations are stored on your server.

  3. 03 · Knowledge

    Document index

    The folders you choose are split into passages and stored in a vector database on the same server. Collections are linked to groups in your Microsoft directory.

  4. 04 · Inference

    Model server

    An open-weight language model, served by an inference engine such as vLLM or Ollama behind an OpenAI-compatible API.

  5. 05 · Hardware

    Your GPU server

    A Dell PowerEdge or HPE ProLiant server with data-centre GPUs in your building, or GPU machines in your own Azure subscription.

How an answer is made

Grounded in your documents.

The model doesn't answer from memory. It first finds the passages that matter, reads them, and replies with sources you can check.

  1. QuestionA colleague asks something in the chat.
  2. Access checkOnly collections this person may see are searched.
  3. SearchThe question becomes an embedding and is matched against your passages.
  4. ReadThe best passages go to the model together with the question.
  5. AnswerA reply with links to the exact sources.
Example

What is our notice period for supplier contracts?

Standard supplier contracts can be ended with three months' notice, unless the contract says otherwise. Framework agreements signed after 2024 use six months.

Purchasing policy 2025.pdf · p. 4Framework template v3.docx · §12

Models and hardware

Sized to your team.

We pick an open-weight model for your work and size the server to it. Models can be swapped later without anything changing for your team.

TeamTypical modelGPU memoryGood for
Up to 25 peopleMid-size model, about 20 to 30 billion parameters48 GBChat, summaries, questions about documents
25 to 100 peopleLarge model, about 70 billion parameters, compressed96 to 192 GBHarder reasoning, many people at once
Specialist workSeveral models side by sideOn requestCode, translation, very long documents

Indicative. Final sizing follows the discovery call and a short test with your own documents.

On-premises

A server in your building

Dell or HPE hardware that your company buys and owns. We install it, harden it and look after it. The model itself needs no internet connection.

Your Azure

In your own cloud subscription

Already on Azure? We deploy GPU machines in your subscription, in an EU region. You pay Microsoft directly; nothing runs in an account of ours.

Integration and operations

What your IT team will ask.

Sign-in
Microsoft Entra ID over OpenID Connect. Two-step login and Conditional Access stay in your hands.
Documents
SharePoint, OneDrive and file shares you choose, indexed on the server on a schedule you set.
Network
No inbound ports open to the internet. Our maintenance access runs over a VPN you can switch off.
API
An OpenAI-compatible endpoint on your network, with a separate key per application.
Logging
Sign-ins and admin actions are logged. You decide how long chat history is kept.
Updates
Monthly patches. New model versions are tested on your own documents before rollout.
Backups
Encrypted daily backups of settings, index and chats, with restores that we test.
Hand-over
Everything documented. If you stop with us, your IT team can run it themselves.
Use your model from your own tools
curl https://ai.yourcompany.local/v1/chat/completions \
  -H "Authorization: Bearer $APP_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "company-assistant",
    "messages": [
      {"role": "user", "content": "Summarise ticket 4471"}
    ]
  }'

The request never leaves your network.

Let's see where AI saves you time.

A free 30-minute call. Honest advice, even if that means not working with us.

Book a free discovery call