Trust Center

Bep is a safe place to practise. A student has a conversation with an AI actor, sees the result on their own screen, and that is where it ends. We built the tool so that practising feels low-stakes, because that is how you learn to handle a difficult conversation: by trying, failing and trying again without an audience. Privacy is not a setting we added later. It is the principle the product was designed around.

• We do not store conversations. The transcript of a practice conversation is not kept on our servers. It only exists on screen during the session and is gone afterwards.
• We do not share results with lecturers. There is no lecturer dashboard showing what a student said or how they performed. Practice is for the student, not for assessment.
• Feedback is not kept either. The coaching feedback after a conversation is generated on the spot and shown to the student. It is not stored.

What we do store is the minimum needed to run the service: accounts, the organisations they belong to, and the scenarios and training pages that lecturers create. Not the practice itself.

Bep uses Google's language models (Gemini), and for some voice conversations OpenAI's (GPT-Live), for three things: playing the AI actor during a conversation, generating coaching feedback afterwards, and helping lecturers build scenarios.

Two things are important to be clear about:

• Not stored is not the same as not processed. To hold a conversation, the words you type or speak have to be sent to the AI model to generate a response. That is processing in transit. We do not store that content, but it is processed by Google's or OpenAI's AI service during the conversation.
• The role-play model and the feedback model are separate. The model that plays the actor and the model that assesses the conversation are two separate calls. The assessment works on the text of the transcript, not on your voice recording.

• Text (typed conversations, feedback, scenario builder): processed within the European Union (the Netherlands, Google Vertex AI europe-west4). This data stays within the EU.
• Live voice conversations: processed in the United States, as a transient stream that is not stored. This applies to Google's Gemini Live and, if your organisation has switched on that option, to OpenAI's GPT-Live.

Why voice is (still) processed in the United States

We are open about this because it is a genuine limitation, not a preference. The natural, real-time voice model we use is currently only usable in the United States. The European versions of this particular voice technology do not yet reliably support multi-turn conversations, and that is exactly the core of a conversation trainer. OpenAI does offer a European endpoint for GPT-Live, but it is not yet available for Bep. This is a limitation of what the AI provider offers today, not something we can design away.

Our position is honest: if processing live voice in the US is not acceptable for your organisation, the voice feature is not the right choice today. Use text mode instead, which is processed entirely within the EU. We re-evaluate this regularly and will move voice as soon as a European model can handle it reliably.

These parties process data on our behalf. None of them receives conversation content other than as described below.

• We do not store your conversations or your feedback.
• Lecturers cannot see what students have practised.
• Text AI runs in the EU (the Netherlands). Live voice AI runs in the US, as a stream that is not stored.
• We tell you exactly what happens to your data, including the part we cannot solve yet.

For the full, legally binding text (retention periods, your rights under the GDPR, how to lodge a complaint), see our privacy statement.