Academy
7. Your Data, Memory and Privacy
Know that models do not learn from your conversations, that data is sent only for the current request, and that sovereign model options exist.
When you chat with an AI model, it might feel like it's learning about you — remembering your preferences, getting smarter over time. In reality, a model is fixed once it has been trained. Your conversations do not change it. Talking to it does not teach it anything new, and your messages are not folded back into its training data. So why does it seem to improve? Because OUPI can feed relevant context back into each request: your saved memory, your skills, your documents. The model itself stays the same; it's the information surrounding your prompt that gets richer.
Every time you send a message, OUPI packages what the model needs — your prompt, the conversation so far, any attachments — and sends it to the provider of the model you chose. Only the content required for that single request is transmitted. It is not stored for training, not shared with other users, and not used to improve anyone else's model. Once the answer comes back, the provider has no further use for your data. Think of it like a phone call: the operator connects you, the conversation happens, and nothing is recorded for future callers.
Data residency can matter — for legal, regulatory or simply trust reasons. OUPI offers sovereign and European model options so that your data can stay within jurisdictions you're comfortable with. On top of that, your organization's administrator can restrict which models your team is allowed to use. This means that even if a powerful non-European model is available on the platform, your admin can ensure sensitive projects only use approved, sovereign alternatives.
Because the model's memory is limited to the context window — the text it can consider for one request — it literally cannot carry information from one session to the next on its own. When the window is full, the oldest parts are dropped. Continuity in OUPI comes not from the model remembering you, but from OUPI's memory and document features reinjecting the right information each time. Your data powers your experience, not someone else's model.
If data residency is important for your project, check with your administrator which models are approved for your team. Sovereign or EU-hosted models let you keep data within trusted jurisdictions without sacrificing capability.
Attach only the documents your task actually needs. Everything you include is sent to the model provider for that request, so keeping attachments focused protects both your privacy and your token budget.
Three things to remember: (1) Models are fixed after training — your conversations never change them and are never used to retrain them. (2) Your data is sent only for the current request, under OUPI's business terms, and is not stored or shared. (3) Sovereign and European models exist for when data residency matters, and your admin can restrict model choices to keep your team compliant.