Setup is covered on the Mistral page. This post is about why organisations pick Mistral specifically, and whether that reason applies to you.
Why organisations choose Mistral
Mistral is a French company and positions itself explicitly as a European alternative. For some organisations that is the deciding argument: if policy or a tender requires you to show your processor sits in the EU, it saves a discussion.
Combined with Ledger Botje that lines up: the data cache sits at Scaleway in the Netherlands and France, and the data processing agreement names both sub-processors. More on that on the security page.
What that means in practice
Note that the argument has two layers you should keep apart. The integration itself - Ledger Botje - runs in the EU and does not use your data for training. What your AI assistant does with the conversation is a separate choice by that provider.
If you pick Mistral because you want everything inside Europe, it is worth checking both layers rather than assuming one covers the other.
Sample questions
- "Which invoices have been outstanding for more than 30 days?"
- "Show the stock of item A1001 per warehouse."
- "Which orders for customer Jansen have not shipped yet?"
What you will run into
Le Chat works differently from the API. If you use Mistral through Le Chat, the way you connect differs from a custom integration on their API. The setup page shows which route you need.
Be concrete. As with other assistants, sharply phrased questions produce better answers. Name periods, customer names and item codes.
Check what you pass on. Even with a European provider it still holds that you verify amounts before acting on them.
Where to go next
For worked questions: the prompt library. For weighing this against other routes: the comparisons.
Keeping the two layers apart
This is where most of the confusion sits in practice, so it is worth making concrete.
Layer 1: the integration. Ledger Botje reads your Exact Online data, keeps an encrypted cache of it and exposes that as MCP tools. That cache sits at Scaleway in the Netherlands and France. Your data is not used to train models.
Layer 2: the assistant. The AI assistant calling the tools processes the conversation. Where that happens and what is done with it is decided by that provider - not by us.
If you pick Mistral for the European argument, that covers layer 2. Layer 1 was already inside the EU regardless of which assistant you use. Conversely: pick an American assistant and your administration cache does not move there with it.
What you show an auditor
If you have to substantiate this, you need two documents. For the integration that is the data processing agreement, which you generate online yourself; it lists the sub-processors and the retention periods. For the assistant it is that provider's policy.
That looks laborious, but it saves an argument later: you can point per layer at who processes what, instead of having to explain that "it is all European".