You know Proton didn’t make their own LLM, right? They are using multiple different 3rd party LLMs in the background to answer your questions in Lumo. What’s end-to-end encrypted is only the chats between your device and Lumo and your chats stored in Lumo, but those same questions forwarded/sent between Lumo and the LLM are not end-to-end encrypted. Even Proton says it themselves in their blog: Lumo security model: How Proton makes AI private | Proton , saying “So we need to provide the LLM with cleartext messages.” So no, your conversations with Lumo are not only known to you, the LLM needs to read them in order to answer your questions and have contexts to the conversation with you. Homomorphic encryption is how Lumo can truly be private, but that idea has never worked out in practice and does not exist in any practical applications, let alone in Lumo. Of course Proton wouldn’t tell you this upfront in their marketing; they have a lot of half truths in their marketing if you look a little longer, and it’s kind of disgusting of Proton as company preaching transparency.
There are a number of AI models that run fully local on your device or self hosted that are so much better than Lumo in terms of privacy, since your questions and data do not go anywhere else.
The only part that I agree with you is that LLM shouldn’t be a priority at this moment before other features, but I have no place to make that decision. In fact, chatting with LLM via the upcoming chats feature is already being tested internally by Anton and the team.
Read this post of mine below to know my stance.