From the transcript:

As for my final analysis of the situation, for the record, this is not the first thing I would have asked Proton to do. I think there’s a lot of other cool things they could have done, and I frankly would have just preferred if they improved some of the integrations between their current products and just improved on those. But this is fine. I think it’s so far fine. I don’t know if I’m going to be using this, especially because it doesn’t really integrate with anything else too much in the Proton ecosystem. One thing you need to understand about Lumo, which is maybe going to make it or break it for you, is a lot of people use AI as part of their process when it comes to searching for something. And Proton doesn’t have a search engine, and this is something you have to actively go to to use, whereas something like DuckDuckGo’s AI agent, which is something I do find myself just accidentally stumbling on using pretty often now, mainly because I’ll do a search, I’ll go through a few web articles, and then I’ll go, oh, okay, that was interesting. I’m curious, like, if I put this information together, like what an LLM will spit out. And then I’ll just go ahead and click.ai. But that use case isn’t quite a thing for Lumo. Like, Lumo, you’re going to have to log in and just go to Lumo just to go ahead and use your LLM. And at that point, you could have used any of the other ones that probably are better performing, almost as good privacy, or you could have just opened up your local program, which probably has better performance as well. So I think the uphill battle, for me at least, is going to be understanding where this could even fit in my workflow without just using it more. It’s gonna be hard for me to know when this is actually better to use than some of the other tools. Either way, I think the major feat that they’re happy with, as well as myself, is their ability to somehow do end-to-end encryption with these kind of AI models. The end-to-end encryption aspect of it is very interesting and is something that hopefully other providers can learn from to build even better AI models.