Lumo is a solid 4/5. I’ve asked questions about numerous topics and it is pretty good, but not the best. I prefer Lumo for the privacy aspect, but you need to make sure that web search is on and specifically ask it to give answers with up-to-date information through web search. One chat was clearly based on outdated information, as it was giving answers that I knew were incorrect. I asked why and I pointed out I have web search on. It then told me that it decides when the chat begins whether it uses web search or not. (I’m not sure if this has been fixed or if it’s known to be an issue.) Make sure that you ask it if web search is enabled prior to any conversations. The latest training is from April 2024, so if you have questions about a person or situation that has changed since then, you will get wildly different answers with old and new information. One time it fact checked me over a person’s recent death, stating that the person was alive when they aren’t. Like with any AI, it’s a good tool, but ask clarifying questions and ask for updated information, especially if you have a question that is relevant to today. Any AI is only as good as the questions asked and its programming.

UPDATE: Proton launched features as promised and directly addressed 2 of my 5 issues. Images are on the roadmap as well. Proton Doc files say they’re coming soon. That just leaves the hallucination problem, which is still prevalent. That being said, Lumo now fulfills most of my needs and, as promised in my original review, I have switched to Lumo Plus. Review changed to 4 stars. Will update again when images are added and/or when hallucinations are addressed. Thoroughly impressed with Proton’s work here!My issues with Lumo are exclusively feature parity. I use ChatGPT quite heavily, and am looking for an alternative option due to recent updates to it. I don’t expect an exact 1:1, but there are some things that, if included, would cause me to swap over to Lumo as my primary assistant:1. Dark mode. Please.2. The ability to attach Proton Docs. I already subscribe to Proton Unlimited and heavily use Proton Drive. This would make the ecosystem far more usable.3. Image file types as supported file types for uploads. I frequently screenshot things and ask ChatGPT about it. I’d love to do that with Lumo.4. Switches, toggles, customization. I’d love to be able to give Lumo custom instructions without having to upload a document for every chat with instructions written in it.5. Hallucinations. Sometimes they’re charming, but they’re often not. Reduced hallucinations would be a big plus.With all that said, Lumo has started in an excellent place. Proton seems to be doing a great job with frequent enough updates and I’ll be watching this closely. And privacy focus on top of that? Chef’s kiss. If this gets to the point where it can cover most of my daily use cases, I’m switching to it immediately.

It’s a nice chatbot/assistant but this can barely be considered AI at its current development stage. For instance, Lumo couldn’t do the basic task of being asked to find some furniture with specific features, sending me tables full of not only broken links (one table had only 2 working links) some even seemed to have completely made up urls. Most concerningly, on the morning of the USA’s strike on Venezuela, Lumo tried convincing me that the event had never taken place, claiming there were no verified sources (it was headline news on an international scale). Even after being given sources from that same morning, it then claimed that the date was incorrect on the source, only after 2 or 3 more queries did the AI update its response. This shows the AI is not fundamentally on a level where it can be a reliable source for any kind of information, even basic questions have the potential to be incorrect. Although the intention of AI is not to think for you, many people do look to AI for knowledge and genuine help. The larger implication of this is that if the AI is this unreliable, there is NO GUARANTEE that it couldn’t accidentally produce incorrect or dangerous information that could lead to a situation where someone is hurt in some way.

During my very first use of this app immediately upon installing, it became apparent that reading responses would be nigh impossible. Why? Because while the keyboard disappeared upon sending my request, the space it took failed to collapse, reducing the space available to display the response to a mere 6 lines of text. Additionally, a button overlayed the response text. Its purpose seems to be to scroll down to the end of the response. Unfortunately it made it impossible to reach the text displayed next to it, for the purpose of scrolling. So the only area that would respond to scrolling exists above that button-a mere three (3) lines of text. Imagine having to scroll through several paragraphs of a response when you can see only 6 lines at a time, and can use but 3 of them to scroll through the remainder. An utter usability disaster, this app does not lend itself to be used for anything serious, it its current state.