Your data, your rules, written down.
An AI agent reads your orders, your customer emails, sometimes your contracts. This page answers what your IT team and your DPO will ask: where the data goes, which models see it, what the agent may do alone and how you stop it.
01
Where your data and models live
You choose, project by project, and the choice is written in the contract.
| Option | Where it runs | When it fits |
|---|---|---|
| EU cloud | Model and data processed in European data centers | Most projects. Quick to set up and GDPR compliant |
| Private cloud | A dedicated environment in your cloud account | When your IT wants everything in its own perimeter |
| Your servers | Open-weight models on your own hardware | Sensitive data that must never leave your network |
02
Which models see your data
We pick the model for the task and for your constraints: European models such as Mistral when sovereignty matters, other providers when the task needs them, open-weight models when data must stay on your servers.
We only use business terms that exclude training on your data, and we show you those terms before the project starts. Each provider says so on its own pages, read on October 1, 2026: OpenAI does not train on data from its business plans and its API by default; Anthropic does not use inputs or outputs from its commercial products, Claude for Work and the API, by default; Mistral opts its Enterprise customers out by default and lets API customers opt out, which we do on every project.
A model hosted on your servers settles the question differently: nothing leaves your network at all.
OpenAI: enterprise privacy Anthropic: is my data used for training? Mistral: training controls Private LLM hosted in Europe or on your servers
03
GDPR, in the design
- A data map for each agent, listing the data it reads, writes and keeps, with the retention period.
- A data processing agreement with you, and the list of subprocessors for your project.
- The agent only reads the personal data its task needs.
- Retention rules set before go-live, and applied automatically.
04
The EU AI Act, handled
Before we build a system, we check which EU AI Act category it falls into and write down what the regulation requires for that category. Human oversight goes into the workflow itself, and anyone dealing with the agent is told it is an AI.
05
Control, logs and the off switch
- Least access. Each agent has its own account and only the permissions its task needs.
- Human approval. Thresholds you set decide what the agent may do alone and what waits for a person.
- Every action logged. The log keeps what the agent read and what it decided, with the reason, for your audits.
- Secrets protected. Keys and credentials are stored in a secrets manager, never in code.
- One switch to stop it. An agent can be paused at any time; the work falls back to your team.
06
The demos on this site
The live demos run on fictional data. What you type is sent to a language model to produce the answer and is not stored by us. To limit abuse, we keep daily counters tied to an anonymous identifier derived from your connection, which changes every day and is deleted after two days.
Send us the questions from your IT team or your DPO. We reply within one business day.
Send your questions