Encryption at rest, no training on your data, one-click deletion, and open-weights models. Here is the whole picture, including the limits.
Plenty of AI products say "we take your privacy seriously" and leave it there. This page is the specific version: what Berges AI does, and where the honest boundaries are.
Nothing is stored until you save a conversation. Once saved, every message and the title are stored as AES-256 ciphertext, so on disk your conversations are not human-readable. This is at-rest encryption, not end-to-end: a message is decrypted in memory to generate a reply, then stored back as ciphertext.
Your chats are never used to train a model. A conversation is stored only when you save it, so you can return to it, and for nothing else.
You can permanently delete any saved conversation with one click. Keep what you want, remove what you do not.
The business model is the subscription. We do not sell your data and there is no ad profile. You are the customer, not the product being sold to someone else.
The underlying models are open weights, published by labs like Moonshot, DeepSeek, and Z.ai. The model is auditable rather than a sealed box.
We do not claim end-to-end encryption, zero-knowledge storage, or that your data never leaves a machine. Saying so would be marketing, not fact. We would rather you trust the specifics than a slogan.
Nothing is stored until you save a conversation. Saved conversations are encrypted at rest, never used for training, and deletable in one click. It is not end-to-end encrypted, and we say so plainly.
It is a paid subscription. That is the trade: you pay, and your data is not the product.
We do not sell your data or build an ad profile from it. Saved conversations are stored encrypted at rest so you can revisit them.