Lumo 2.0 Review: Proton's Private AI Assistant Grows Up
Proton rebuilt Lumo, its no-logs AI assistant: Lumo 2.0 is multimodal, much faster, searches the live web with citations, adds memory and a business tier — and still keeps every chat and image zero-access encrypted. Our review.
Table of contents
Most AI assistants ask you to make a quiet trade: you get a capable chatbot, and in return your conversations become data — logged, analyzed, and often used to train the next model. Proton's Lumo was built to refuse that trade. Its original pitch in 2025 was simple: an assistant that does not log your chats or train on your data, backed by Proton's Swiss privacy pedigree. Over 10 million people signed up. Lumo 2.0 is the first major rebuild since launch, and it is a much bigger jump than a point release usually implies.
What actually changed
The headline is capability. Proton says Lumo 2.0 Max scores 240% higher than the previous version on an independent benchmark spanning coding, scientific reasoning, and general knowledge, with everyday responses up to 76% faster. Complex tasks now show a visible "thinking" state, so you can follow the reasoning as it happens rather than staring at a spinner. Those figures are vendor-reported, so treat them as direction-of-travel rather than gospel — but the practical experience of a faster, more coherent assistant is easy to feel in normal use.
It is now multimodal
The biggest functional gap in the old Lumo was that it only did text. Lumo 2.0 closes it:
- Analyze images you upload — screenshots, documents, photos.
- Generate visuals from a text prompt or a rough sketch.
- Edit existing images inside the same conversation.
Crucially, Proton's zero-access encryption extends to images too. By design, neither the images you upload nor the ones Lumo generates are accessible to anyone — including Proton itself. That is the differentiator: multimodal AI where the visual data is encrypted rather than mined.
Web search that cites its sources
When a question needs current information, Lumo 2.0 pulls live results and cites where they came from. It can follow recent news, live financial data, and weather, and Proton says it hallucinates less than the previous version on these queries. Sourced answers matter more for a privacy assistant than most: if you are not going to hand your data to a big-tech model, you still want the answers you do get to be verifiable.
Deeper memory and context
The context window is now twice as large, so Lumo holds more of a long conversation or document in view at once. Memory lets it learn your preferences and working style over time — and, in keeping with the privacy framing, you control exactly what it retains or forgets. Projects keep related chats, files, and instructions bundled together across sessions, which is the feature that turns a chatbot into something you can actually run a piece of ongoing work through.
Lumo for Business
New in 2.0 is a business tier aimed at organizations that cannot accept the usual AI risk profile: employee prompts quietly becoming training data, confidential documents exposed, or infrastructure sitting under US legal jurisdiction. Lumo for Business keeps every conversation zero-access encrypted, never logged, never used for training, and hosted on European infrastructure. For regulated industries, legal teams, and anyone handling sensitive material, that combination is the entire sales pitch — and it is a genuinely different one from the mainstream assistants.
The honest caveats
- The benchmark numbers are Proton's own. A 240% jump is impressive but self-reported; independent head-to-heads with the frontier models (GPT, Claude, Gemini) will tell you where Lumo really lands on raw capability.
- Privacy has a ceiling of usefulness for some workflows. Because Proton cannot see your data, some conveniences other assistants offer by processing your history server-side are handled differently or left to you to control.
- It is a privacy-first assistant, not necessarily the single most powerful one. If your only metric is benchmark-topping intelligence regardless of data practices, the frontier labs still set the pace. Lumo's argument is that for most everyday work, it is now capable enough — and it is the only one of them that does not read your conversations.
The takeaway
Lumo 2.0 is the version that makes the privacy promise easy to accept rather than something you tolerate for the principle. It is faster, meaningfully smarter, multimodal, and it finally searches the live web with citations — while keeping the one guarantee its rivals cannot match: your conversations and images are encrypted so that not even Proton can read them. For anyone who has wanted to use an AI assistant without feeding a data machine, this is the first release where you no longer have to give up much to get it.
