Proton Lumo: A Privacy-Focused AI Chatbot That Aims to Compete Proton released its privacy-focused chatbot, Lumo, in mid-July. Following an update in August that improved its performance, I find myself using it more than ChatGPT or Claude. In a world where tech companies often exploit user data, I am actively seeking more ethical tools. But with competitors offering flashy features in exchange for personal information, can Proton seriously hope to compete? Eamonn Maguire, Proton’s head of machine learning, is confident. He believes Lumo has found a valuable niche. The project began last year after the positive reception to Scribe, Proton’s AI email writing tool. The success of Scribe changed how the company viewed AI. Historically, Proton has been late to market with its products, launching Mail a decade after Gmail and Drive eight years after Google Drive. They were determined not to be late to the chatbot revolution. Maguire states that moving quickly was essential to address a looming privacy problem. He points to the monetization path of platforms like Gmail, which turned to ads and selling user data. He argues this pattern is repeating with AI chatbots, citing Elon Musk’s plans to introduce ads in Grok’s responses. The enshittification of AI, as he implies, has already begun. Maguire believes Proton can compete for two key reasons: the rising quality of open-source models and the fact that most users do not need the most powerful AI. Open-source models, particularly from China like Zhipu AI’s GLM-4.5, are now competing with proprietary ones on performance benchmarks. Lumo uses a combination of smaller, efficient open-source models, including Nemo, OpenHands 32B, OLMO 2 32B, and Mistral Small 3. This approach allows Proton to be nimble and cost-effective. Maguire compares it to driving a Formula One car to the grocery store; it is overkill. For most tasks, a smaller, capable model is sufficient. This efficiency lets Proton offer Lumo at a lower price. A subscription costs $13 per month, undercutting the $20 basic plans from competitors and far less than the new $200 unlimited tiers some are introducing. Lumo also benefits from being integrated into Proton’s existing ecosystem of encrypted email and cloud storage, gaining context from the user’s digital environment. Furthermore, chatbots are tools without strong network effects. A user might use Claude for coding but switch to another service for a different task, as no single platform is best at everything. Despite these advantages, the challenge is immense. I think of Mozilla’s Firefox, a superior privacy-focused browser that holds only a fraction of Chrome’s market share. There is also the AGI-sized elephant in the room. Major AI companies are engaged in a costly race to build artificial general intelligence, a contest with potentially only one winner. How can a small player like Proton compete? Maguire’s answer returns to the car metaphor. You do not need a spaceship to run errands. If the goal is to help people be productive and learn, AGI is probably unnecessary. Proton is not under the illusion that everyone will switch from ChatGPT to Lumo. Their goal is to provide the best ecosystem for doing things within a privacy-preserving framework. With 100 million users for Proton Mail, the company has already proven that a privacy-first service can find a substantial audience. It is competing in a space that some claim requires trillions in investment. If Proton can demonstrate that AI does not have to be the antithesis of privacy, that alone could make Lumo a success.
