When Proton — the Swiss company best known for Proton Mail and Proton VPN — launched Lumo in July, it pitched the assistant as a privacy-first answer to the big-tech AI assistants: no secret training on your data, encrypted chats, and an open-source approach that lets independent eyes check the plumbing. Less than a month later, Proton has pushed the first major update, Lumo 1.1, and it’s not just a polish pass. According to the company, this release materially changes what the assistant can do,

Proton’s announcement for Lumo 1.1, posted on their blog on August 21, 2025, reads like a checklist aimed at the two problems every newcomer to the assistant space must solve: match the raw capabilities of the incumbents, and keep user data private while doing it. In the blog post, Proton says Lumo 1.1 is “faster” and “smarter,” and that the update brings substantial gains in a handful of measurable areas.

Those gains are impressive on paper. Proton claims Lumo 1.1 delivers:

• 170% improvement in context understanding (better answers from documents and uploaded files),
• 40% better at producing working code, and
• 200%+ improvement when solving multi-step problems and choosing the right internal tools — including web search.

Taken at face value, those numbers would move Lumo from “promising privacy experiment” toward “practical everyday assistant,” especially for power users who feed an assistant documents, ask it to troubleshoot code, or expect it to carry state across a long, multi-part task.

That said, Proton hasn’t published a detailed methodology showing how it measured those percentage gains. The company shared comparative screenshots and examples that highlight clearer, better-formatted answers from Lumo 1.1, but it didn’t open the entire testing suite to independent verification in the blog post. That omission is important: high-level performance claims are useful headline copy, but they mean more when the tests and datasets are public.

Proton did, however, release the Lumo security model and the mobile app code — a move aimed at building trust. The security writeup explains the company’s “zero-access encryption” approach (how saved chats are encrypted and stored so even Proton can’t read them) and the release of mobile source code lets outside researchers audit the client side. Those are strong gestures for a company that’s trying to make privacy its competitive advantage.

One of the most tangible upgrades in Lumo 1.1 is its web-search integration. Early versions of many assistants stumble on current events or on prompts that require looking something up; Proton says Lumo 1.1 has a better search tool and that the assistant now “suffers from fewer hallucinations” when addressing current events. If true, that’s a big step: web access is where many assistants make or break a user’s trust.

But again: the company’s write-up shows examples rather than a full error-rate analysis, and independent testing by third parties will be the acid test for whether Lumo’s search-driven answers are reliably accurate.

Lumo is available via web and mobile apps. Proton rolled Lumo 1.1 out to both free users and paying subscribers — free accounts still have usage caps, while Lumo Plus subscribers get higher or unlimited usage and some advanced features. The mobile apps have also been updated to point users at the new models.

Proton’s pitch isn’t only about better answers — it’s about offering an alternative business model. Where many mainstream assistants are tightly integrated with ad ecosystems or training pipelines that harvest user content, Proton repeatedly stresses that Lumo does not keep logs, does not train on user conversations, and stores saved chats with encryption that prevents Proton itself from reading them. Releasing client code and the security model is part of a transparency play: invite the community to check the work and help build credibility.

That approach raises interesting tradeoffs. Running and fine-tuning large language models privately and on Proton’s infrastructure — while avoiding harvesting user data — is expensive. Proton’s challenge will be sustaining high performance and availability without compromising privacy or forcing a heavy paywall on users who value those protections.

Lumo 1.1 is more than a bugfix release: it’s Proton staking a claim that privacy and performance don’t have to be mutually exclusive. The numbers are headline-grabbing, the open-sourcing and security documentation help Proton’s credibility, and the upgraded search and multi-step reasoning make the assistant more useful. The missing piece is independent verification of the percentage gains and long-term proof that Proton can scale that model without trading away its privacy guarantees. For now, Lumo 1.1 is the kind of bold, still-young product that’s worth trying if you prioritize confidentiality — and worth watching if you care about where the AI assistant market goes next.