Lumo makes the stronger encryption promise. Berges AI makes the stronger output promise. Both run open-weights models and refuse to train on your chats.
Proton built Lumo the way it builds everything: zero-access encryption, so Proton itself cannot read your chat history, open models run on its own European servers, and open-source apps you can inspect. Since its 2025 launch it has added a thinking mode, memory, projects, and image generation.
Berges AI encrypts at rest rather than zero-access, which we say plainly, and puts its effort into how the model answers: a frontier open-weights reasoning model on every turn, interceptors, and no filler. Two honest products with different centers of gravity.
Lumo's zero-access encryption means the stored history is unreadable to Proton. Ours is unreadable on disk but decrypted in memory by our servers to build each request. Lumo's is the stronger guarantee for stored history. In both products the model still has to see your message in plain text to reply, and in both it is not used for training.
Proton runs its models on hardware it controls in Europe. We store conversations on EU servers and send each request to Fireworks AI, which states it keeps nothing after the reply. If keeping inference inside one company's walls matters to you, that is a point for Lumo. If you want a wider choice of large open-weights models, that is a point for us.
Lumo fits naturally if you already use Proton Mail, Drive, and VPN. Berges AI is one thing: a chat that answers directly, runs a reasoning model on every turn, and lets you shape responses with interceptors. Neither needs the other.
For stored history, yes: zero-access encryption is a stronger promise than encryption at rest, and we do not pretend otherwise. Both products run open-weights models and refuse to train on your chats. Lumo keeps web search off by default; Berges AI searches through its WebSearch interceptor when the model judges a question needs it, and shows every search on the reply.
Both run strong open-weights models. The difference you will notice is style: Berges AI is tuned to answer first and skip the padding, and every reply comes from a reasoning model that shows its working. Lumo has a thinking mode you switch on.
Lumo Plus is cheaper than Berges AI Pro. Berges AI has a $5 plan for light use. Both have limited free tiers.