How we use AI
Smart where it helps. You stay in charge.
You decide, field by field, what a model may see. Every step is recorded, and that helps you show you meet the GDPR, ISO 27001 and the AI Act.
your data first, then the model
Last checked: September 25, 2026 We re-check the providers, countries and controls on this page at fixed intervals.
Where it gets smarter
Four places where your software saves work
Fixed rules and your own data do most of it. A model only joins in where it really adds something, and then for one small task.
No double work
You type something once, and it is right everywhere.
- How it works
- Fixed connections and rules pass the data on. A model only reads what has no fixed shape, such as a PDF, an email or a photo of a receipt.
- For example
- work order to invoice, price list into the quote, intake into the file
Where does it get stuck
You see where work is waiting, and for how long.
- How it works
- Every change is recorded: who, what and when. Fixed rules work out the waiting times and lead times from that.
- For example
- quotes waiting for approval, reports not yet signed off
What changed since last time
A short summary per file of what changed since your last visit.
- How it works
- Rules determine exactly what changed. A small model turns that into one readable paragraph, using only the fields you release.
- For example
- care plan, customer file, project, contract
Recurring work
The weekly routine runs by itself, and a person approves it.
- How it works
- A fixed schedule and workflow do the work. Nothing goes out without approval.
- For example
- order proposals, reminders, weekly reports
How a smart step works
The model is the smallest part
Your data in order first, then fixed rules. Only then does a model get one clear task. That is cheaper, faster and more predictable.
Your data
In one place, sorted field by field: what is public, what is internal and what is confidential.
Fixed rules and tools
Counting, comparing, sorting and linking are done by ordinary software. Always the same answer.
A small task for a model
Only where it adds something: reading a document or writing a summary. With only the released fields.
Check
Fixed rules check the result, and a person decides before anything changes.
Recorded
Which fields, which model, where it ran, who approved it and when. You can look it up in the log.
Where a model runs
You choose where it runs
We show the company’s country and where you can run the model. You choose for each smart step.
Claude
- Company
- Anthropic, United States
- Runs
- At Anthropic, in the US
Strong at reading, summarising and making proposals.
We set it up for youOpenAI
- Company
- OpenAI, United States
- Runs
- At OpenAI, in the US
For searching by meaning and reading documents.
We set it up for youMistral
- Company
- Mistral AI, France
- Runs
- Through a selected European inference endpoint
A European provider, for when the EU is your first choice.
We set it up for youOpen models via Ollama
- Company
- The open model you choose
- Runs
- On your own server or with a host in the EU
The model runs on a server you choose. Nothing goes to a model provider.
We set it up for youGPU by the hour in the EU
- Company
- RunPod, United States
- Runs
- In a data centre in the EU
Computing power by the hour for heavy work, such as making a 3D model. It is switched off again when the job is done.
We set it up for you
This website · Mistral AI, France. Understands your question on the start page, after contact details are removed. The default API does not guarantee a processing location.
What never goes to a model
What a model never sees
A model only gets what that one task needs. This never goes to it:
- Passwords, keys and the login details of connections.
- Fields you shield. You shield names field by field.
- Data you mark as strictly confidential.
- More than the task needs: only the released fields and rows, never your whole file.
Smarter, but never out of control.
For every smart step you see five things, and you can always switch it off.
- Which fields it sees. On or off, field by field.
- Where it runs. Region and company, by name.
- Who approved it. A smart step only goes on after your administrator agrees.
- The trail. What it did, when and for whom, visible in the log.
- Switch it off. One switch, and the work carries on with fixed rules.
- Fields
- Goals, agreements, dateName shielded
- Runs at
- Mistral AI, France
- Approved by
- Administrator
- Recorded
- Every time: who, when, which fields
That is how it helps you show you comply: with the GDPR (no more data than needed), ISO 27001 (logging, access, suppliers) and the AI Act (transparency and human oversight).
Your own assistant
Do you work in ChatGPT, Claude, Mistral Vibe or Copilot?
Read how Studio works with your own AI assistant, what it deliberately may not do, and why.
Pilot first
We deliver the pilot, so you know what to ask for
One concrete problem in your own work, small and bounded, with a result agreed up front. Then you choose: carry on with us, or go to the market. Both are fine.
Where could it be smarter for you?
We look at your work together and show where a smart step saves time, and where ordinary rules are enough.