Private AI chat: why your conversations should never train a model

People tell chatbots things they would not tell many humans: health worries, money problems, drafts, ideas, code. That makes privacy a core feature, not a bonus.

What can happen to your chats

Depending on the provider, a conversation may be:

• Stored for a long time, often with your account name attached
• Read by staff for quality or safety review
• Used for training, which means pieces of it can influence future models

Training is the part most people overlook. Once data went into a model, you cannot take it back. Rare but real cases exist where models repeated fragments of their training data.

What a private AI service should promise

1. No training on your data. Not now, not later, and not through partners.
2. Encryption at rest. If the database leaks, messages are unreadable.
3. Easy export and deletion. Your data, your decision.
4. No trackers or ads. A chat page does not need them.
5. Clear legal basis. A real imprint, a privacy policy, and a company you can contact.

How Orrevan handles this

• Orrevan does not train on customer data.
• Chats are stored encrypted.
• You can export and delete your data.
• The site uses no ad trackers and no third-party analytics. Only a session cookie is needed.
• The operator is Endemine OÜ in Estonia, and the legal texts are public: Privacy, Terms, Imprint.

Privacy is also about control. An uncensored model that you can only use under strict logging would defeat the purpose. That is why the privacy rules and the model choice belong together.

Create an account and see how it feels to use AI without being the product.