Berges AI

Berges AI is built for clear, efficient help—without the extra chat. Instead of forcing everything through one mode, it uses a three-layer approach so quick questions don’t slow you down, while harder problems can trigger deeper reasoning.

The fast-response layer handles the common 90% of questions, and the deep-thinking layer steps in automatically for tough requests (or when you explicitly ask for “think hard”). You can also choose a Berges interceptor—Reasoning, Brainstorm, or Empathy—to shape how the answer comes out.

Your conversations stay yours. Messages and conversation titles are stored as AES-256 ciphertext, conversations aren’t used for training, and you can delete any conversation in one click. Berges AI runs on open-weights models under Apache 2.0 or equivalent licenses.

What does the “three layers” approach mean?

Berges AI cascades work across a fast-response layer for easy questions, a deep-thinking layer for harder ones, and an interceptor layer to shape the style of the output.

When does the deep-thinking model kick in?

It runs automatically for hard questions, and you can also summon it on demand by saying “think hard.”

What are the Berges interceptors (Reasoning, Brainstorm, Empathy)?

They control how the answer is produced, letting you steer the response toward reasoning-focused, idea-generating (brainstorm), or people-oriented (empathy) outputs.

Do you train on my conversations?

No. Conversations aren’t read and aren’t used to train any model.

How is my data stored, and can I delete it?

Conversation titles and messages are stored as AES-256 ciphertext. You can delete a conversation anytime with one click.

What pricing plans are available?

Plans are Basic ($5/month), Pro ($19/month), and Plus ($99/month), each with different usage limits and suited to light, everyday, or heavy use.