With Lumo, Proton thinks it can carve a place at the AI table (www.engadget.com)

Proton quietly launched Lumo, a privacy-focused chatbot built from an ensemble of open-source models (Nemo, OpenHands 32B, OLMO 2 32B, and Mistral Small 3) and iteratively improved after an August patch. Positioned inside Proton’s existing apps (Mail, Drive), Lumo offers a free tier and a $13/month paid plan to lift rate/token limits — deliberately cheaper and lighter-weight than the high-end offerings from OpenAI and Google. Proton argues smaller, composable models can deliver “capable” responses for everyday productivity without the compute, cost, or data-harvesting tradeoffs of state-of-the-art proprietary systems. For the AI/ML community this matters because it shows a viable path for privacy-preserving products that rely on the rapidly improving open-source model landscape. Benchmarks are clustering and some open systems (e.g., GLM-4.5) are reaching top ranks, making leaner stacks more competitive. Proton’s approach highlights engineering and product trade-offs: ensemble smaller models to reduce resource and business risk, focus on integration and user privacy rather than chasing AGI, and target practical workflows where immense model capacity is overkill. The experiment tests whether a sustainable, non–data-extractive chatbot can retain a meaningful niche amid ad-driven, feature-heavy rivals — a practical case study in the economics and ethics of deploying LLMs.