Every message and conversation title is stored as AES-256 ciphertext. Here is what that means, in plain terms, including what it does not cover.
Most AI chats store your messages as plain text in a database. Anyone with access to that database, by accident or otherwise, can read them. Berges AI stores every message and conversation title as AES-256 ciphertext instead. At rest, on disk, your conversations are not human-readable.
We would rather explain the limits of that than oversell it, so here is the honest version.
When a message is saved, it is encrypted before it touches the database. What lands on disk is ciphertext. If someone got a copy of the raw database, they would see scrambled bytes, not your conversations. Conversation titles are encrypted the same way.
This is not end-to-end encryption. To answer you, the model has to read your message, so it is decrypted in memory to generate a reply, then stored back as ciphertext. We are telling you that plainly because a lot of products imply more than at-rest encryption actually provides.
Your conversations are never used to train a model. They exist so you can come back to them, and for nothing else. You can delete any of them permanently with one click.
The models Berges AI runs are open weights, published by their labs. That does not encrypt anything, but it does mean the model itself is auditable rather than a sealed system you have to take on faith.
No. Messages are encrypted at rest with AES-256, but they are decrypted in memory to generate a reply. We do not claim end-to-end encryption, because it would not be true.
We do not read your conversations and we do not train on them. They are stored encrypted at rest so they are not casually readable from the database.
It is permanently deleted. Deletion is one click and it is not a soft hide.