If you’ve been side‑eyeing the big consumer chatbots because you don’t love the idea of your drafts, travel plans or client notes sitting on someone else’s servers, Proton wants you to know it sees you. With the latest Lumo 1.3 update, the company is turning its “privacy‑first AI” pitch into something more concrete: a proper encrypted workspace for long‑running projects, not just one‑off prompts.

Instead of dumping everything into a single history and hoping you can scroll back to that one brilliant idea from last week, Lumo now lets you spin up dedicated Projects — self‑contained spaces where you can park a semester‑long research paper, a side‑hustle roadmap or that over‑ambitious three‑country trip you swear you’ll actually take this year. Each project keeps its own running memory, so the chatbot remembers the context, files and links tied to that work without you having to re‑explain yourself every time you open a new chat. On a practical level, that’s the difference between “help me outline chapter three of the same paper you’ve seen for weeks” and “hi, here is my life story all over again.”​

The privacy bit isn’t just branding either; Projects are described as their own encrypted environments, and Proton says chats and files stored there aren’t used to train AI models. That’s a deliberate contrast with most mainstream assistants, where you often have to dig through settings or opt‑out flows to limit training, and even then, you’re still trusting a company whose business model depends heavily on data. Proton is leaning on the same posture it built with Proton Mail and Proton Drive — end‑to‑end encryption by default, minimal data collection and an explicit promise that your content is off‑limits for model training. For journalists, lawyers, researchers or anyone working with sensitive material, that alone will make Lumo feel more like a working tool and less like a toy.​

Projects also hook directly into Proton Drive, so the AI isn’t just chatting with isolated uploads; it can reference documents and other files in your encrypted cloud as part of the same workspace. You can throw PDFs, drafts and supporting material into a project, then ask Lumo to summarize, compare, rewrite or fact‑check against that corpus while everything stays tied to the same context. Because those resources sync across devices, you can start outlining on a laptop, refine on a phone and pick up again on a desktop without losing the thread or having to shuffle files around. It’s closer to how people actually work on long projects, especially if you’re juggling multiple roles or clients.

Under the hood, Proton says Lumo is powered by a mix of open‑source models, including Nemo, OpenHands 32B, OLMO 2 32B and Mistral Small 3. That’s an interesting decision in an ecosystem dominated by proprietary giants; it gives Proton more control over where and how models run, and it aligns philosophically with the open‑source security community it’s been part of for years. For users, the takeaway is that this isn’t another “we’ve wrapped GPT-4 in a new skin” product; Proton is curating and orchestrating a set of models inside its own infrastructure, then wrapping that in the company’s familiar encryption stack. Whether that can match or beat the absolute cutting‑edge on raw reasoning is a different question, but the trade‑off is clear: a bit less flash in exchange for more control and privacy.​

There is, of course, a freemium catch. Everyone gets access to Lumo 1.3, but free accounts are limited to a single project — enough to feel the workflow, not enough to live in it. If you want unlimited projects, Proton sells a Lumo Plus subscription for $10 a month, which effectively turns Projects into a paid productivity layer on top of the basic chatbot. For people already paying for Proton Mail or Proton Pass, that’s another line item to consider, but it also nudges Lumo into the same mental bucket as productivity suites and note‑taking apps, not just “another chatbot tab” you occasionally open and forget.​

Zoom out a bit, and this update feels like a quiet but important shift in how AI assistants are being positioned. The first wave was all about raw capability: bigger models, more tokens, more impressive demos. The second wave is about trust and fit. Can you use this thing for client work without violating an NDA? Can your students draft essays there without seeding their ideas into a training corpus? Can a newsroom brainstorm sensitive investigations without worrying that their queries might surface as “hallucinated examples” somewhere else? Proton is betting that for a decent chunk of users, the answer with mainstream tools is still “not really,” and that encryption plus projects is the kind of feature set that could tip the scales.

It also underlines a subtler point: context is where AI becomes genuinely useful, and context is where privacy stakes get higher. The more an assistant remembers about your ongoing work, the more powerful it becomes — but the more catastrophic it is if that data leaks, gets misused, or is repurposed to train models you never signed up to improve. By carving out encrypted, project‑scoped spaces and drawing a hard line against training on that data, Proton is trying to square that circle. It’s not a magic answer to every privacy concern, but it’s a concrete product decision that nudges AI toward something people can actually rely on, not just marvel at.​

For now, Lumo’s Projects are likely to appeal most to people already in Proton’s orbit: privacy‑conscious users, security‑minded professionals and anyone uncomfortable with the “your data makes our AI better” bargain baked into most assistants. Over time, though, this kind of feature may become table stakes. As more of our work, homework and planning quietly route through AI tools, an encrypted space for those projects isn’t just a nice‑to‑have; it starts to look like the baseline we should have been asking for all along.