Where Is Your Data When You Use AI? EU Server vs. Model Calls

BOTrento Team9 October 2026 · Updated 12 October 20269 min read

The data flow in practice: three places where your data lives

Take a typical case: an enquiry arrives in your shared inbox asking for sound and lighting at a company party. The AI is meant to sort it, pull out the date and scope, and prepare a draft reply. Along the way, the data moves between three places.

1. Your source systems

Your mailbox (Outlook, Gmail) and your CRM or spreadsheet (HubSpot, Pipedrive, a shared sheet). This is where the data originates and where it stays. Where it physically sits is decided by your email or CRM provider, not by the AI tool.

2. The AI service's storage

The service that runs the process stores what it needs to do its job: case status, step history, draft replies and quotes, and a record of who approved what. This is the place the sentence "we store data on servers in the EU" refers to.

3. Processing by the language model

When the model has to read an enquiry or write a draft, the service sends the text to the model provider through its API. The model processes the text and returns a result. This step runs on the model provider's infrastructure, not on the service's server.

The key point: your own EU server that calls an external AI API is not a closed computation. Storage can be in the EU, but the text needed for the task is always processed by the model provider. Anyone who claims otherwise should be able to explain exactly where the model runs.

What goes to the model provider and what does not

A well-designed process sends the model only what a specific task requires. The table shows the difference.

How to limit what the model sees

• A narrow brief: "prepare a reply to this enquiry" rather than "go through the whole mailbox".
• Field selection: pull only the CRM fields needed for the given step.
• Sensitive attachments left out: contracts, personal ID documents or health data do not belong in an ordinary sales process.
• One case type: the narrower the process, the easier it is to check what leaves it.

For a closer look at the legal basis and the processor's role, see GDPR and AI in your business: a practical checklist.

Training on your data: chat vs. API

Whether an AI learns from your enquiries depends mainly on which route the data takes to the model.

Self-service chats

When an employee pastes an enquiry into a personal chat account, processing follows the terms and settings of that particular account. The company often does not know who pasted what, or under which settings. That is the main risk: nothing secret is going on, it is simply that nobody is watching. The difference between a chat for individuals and a company process is covered in ChatGPT or Claude for your business, or a managed AI process?

Commercial API

When data is sent through a commercial API, the providers do not use it to train their models. That does not mean the text never exists at the provider, because it has to receive it to process it. So check the chosen provider's terms for whether and for how long inputs are retained, and for what purpose.

Who is the contracting party

If the service calls the model through its own account, the model provider is its sub-processor. If you use your own API key, you have a direct relationship with the model provider, you can see usage and you know who receives your text. The help page on your own API keys explains how this works.

Checklist: 8 questions for an AI tool vendor

Before you let AI near a mailbox full of enquiries, get written answers to these questions:

1. Where is the service's data physically stored (country, hosting provider)?
2. Which language model providers process the text, and through what type of access (API or consumer account)?
3. What data is sent to the model at each step of the process?
4. Does the model provider use inputs for training? Where do its terms say so?
5. Whose API key is used, and can you use your own?
6. Does anything go to a customer without human approval?
7. Who has access to the data, and how is access removed?
8. What happens to the data when the relationship ends (export, deadline for deleting copies)?

Email template for a vendor

Hello,

We are considering your tool for processing sales enquiries and quotes. Could you please clarify in writing:

1) Where do you store our data, and with which hosting provider?
2) Which language model providers process our texts, and through what type of access?
3) What data is sent to the model, and what stays only in your storage?
4) Does anyone in the chain use our data to train models?
5) Can we use our own API key?
6) How are outgoing emails and quotes approved?
7) What do data export and the removal of access and copies look like when we end the relationship?

Thank you,
[name, company]

If you get an answer like "everything is in the EU, you needn't worry" and the vendor does not distinguish between storage and the model, ask again. For specific situations, such as enquiries that contain sensitive personal data, we recommend speaking to a lawyer. The general duties of an AI user are summarised in EU AI Act for small businesses.

How BOTrento handles the data flow

BOTrento Chief of Staff is a managed process for enquiries, quotes and other sales steps. Here is how it answers the checklist:

• Storage: we store service data on servers in Germany (EU).
• Model processing: the language model receives only the texts needed for the task, through the API of the provider you choose with your own key (Anthropic Claude, OpenAI, Google Gemini, Groq, ElevenLabs). We say openly that this is not a closed computation within the EU.
• Training: commercial API providers do not use the data for training.
• Your own API keys: you pay for AI directly to the provider with no mark-up. Keys are encrypted, only the last 4 characters are visible in the interface, and you can remove them at any time.
• Human approval: nothing is sent without approval. Drafts of emails and quotes are approved by your own person.
• Offboarding: you receive an export of data and outputs and a description of the processes. Access and copies are removed by the agreed deadline.

Chief of Staff starts with a 14-day free trial, with no payment card and no API key of your own. The 14 days begin with the first successful upload of your data (an import or a mailbox connection), and you have 7 days to prepare. Because you decide which data you upload, you know exactly which data takes part in the process. A process demo shows how such a process looks step by step. The scope of each plan is on the pricing page.