AI Knowledge Assistants

Private knowledge AI

Knowledge assistants built around your data rules: where data is stored, which model provider is used and who sees what, agreed in writing first.

The problem

Sensitive information

Some documents cannot go into a public AI tool.

Unclear requirements

It is not clear where data may be stored, processed or sent.

What is included

Data location

Where documents and the index are stored, agreed per project.

Hosting and providers

Hosting and model-provider options chosen around your requirements.

Access control

Who may ask what, and which sources each answer may use.

Logging

What is recorded, and who can see it.

An example, step by step

Example · synthetic data A healthcare supplier with strict data rules

Input

Written requirements

Data stays in the EU, no provider may train on it, access is limited to two departments, and every question is logged for a year.

  1. Write the requirements down and agree them before any data moves.
  2. Choose hosting and a model provider that meet them, with the contract terms to show it.
  3. Set up access, logging and deletion, then test that each requirement holds.

Output

A data-flow agreement and set-up

A diagram of where data goes and who can see it, the provider and region chosen, and a test record for each requirement.

How it works

Requirements

We write down your data, access and provider requirements.

Design

We choose the set-up that meets them, and say what does not.

Build and check

The assistant is built and checked against the requirements.

What you receive

A written data agreement

Storage, processing, providers and access, agreed before any document is indexed.

A set-up that follows it

Built and checked against the agreement.

Documentation

What runs where, for your records and reviews.

What it works with

Your requirements

Data location, retention, access and logging, in writing.

Your documents and systems

The knowledge the assistant may use, as for the other knowledge services.

Provider terms

The hosting and model provider contracts that back the set-up.

How we check it works

Requirements tested

Each written requirement is checked: region, access, logging, deletion.

Contract terms checked

Provider terms are read against the requirements before go-live.

Answer quality

Evaluated as for the other knowledge assistants, on questions you confirm.

What to keep in mind

  • Not every requirement can be met by every provider; we say so before building.
  • Security depends on your systems and policies as well as the assistant.

What shapes the scope and quote

  • How strict the requirements are, and whether self-hosting is needed.
  • The volume of documents and number of users.
  • Logging, retention and audit needs.

Frequently asked questions

Can the data stay in the EU?

Yes, by default. Documents, the index and processing stay in the EU/EEA; a transfer outside it happens only when agreed in writing, with Standard Contractual Clauses.

Is our data used to train models?

No. We use AI providers under terms that exclude training on the data we send, and name the provider and model in the agreement.

Is this certified for healthcare or finance?

We do not claim certifications. We build to your written requirements and show how each is met; your own compliance review decides what it needs.

Can we choose the model provider?

Yes. The choice is made against your requirements and recorded, together with the contract terms that support it.

Start a project

Have a business problem AI could help solve?

Tell us what you're trying to improve. We'll help you identify a practical next step.