On Tuesday, June 30, Proton launched Lumo 2.0, adding image generation, memory, private web search and new model tiers to what it calls the biggest upgrade yet for its privacy-first AI assistant, according to 9to5Mac .
The timing matters because Lumo arrived last July as Proton’s answer to mainstream AI assistants: no logs, zero-access encryption, no data sharing, no use of conversations for AI training, and open-source language models. Less than a year later, Proton is trying to close the capability gap without walking back that privacy pitch.
June 30: Proton makes Lumo 2.0 its biggest AI upgrade yet
Lumo 2.0 is now available with three headline additions: multimodal image tools, encrypted personalization features, and upgraded web search with live results and citations.
Proton says the assistant can now analyze, edit and generate images inside the same conversation. The company is also adding user-controlled memory, encrypted Projects, and Custom Lumos, which are meant to make the assistant more useful across recurring workflows.
The search upgrade is equally central to the launch. Proton says Lumo 2.0 has “enhanced web search with live results and source citations,” a feature aimed at making answers more current and easier to check.
That combination shifts Lumo from a privacy-focused chatbot into a broader AI assistant. It now covers the same basic job categories users expect from leading AI tools: text, images, memory, files, projects and web-connected answers.
Proton says more than 10 million people have adopted Lumo “as a private alternative to other leading AI platforms.” The company is also using the launch to push Lumo for Business, aimed at organizations that want AI features without exposing company data.
“Lumo for Business is built for organisations that can’t afford those risks. Every conversation is zero-access encrypted, never logged, and never used to train future models.”
Proton also says company data stays on “independent European infrastructure,” adding that access to Lumo “cannot be subject to US Executive Orders” and user data “is not subject to American data collection requests.”
After last July’s debut, image tools and memory push Lumo beyond text
The biggest product change is visual. Image recognition and generation give Lumo 2.0 a role in creative and analytical workflows that Lumo’s original text-first version could not fully serve.
That matters because Proton’s privacy argument becomes more demanding once users upload images, documents or business material. The more useful the assistant becomes, the more sensitive the inputs are likely to be.
Memory is the second major test. Proton describes the new feature as user-controlled, with encrypted Projects and Custom Lumos designed for “more personalized and productive AI workflows.”
In practice, that means Lumo 2.0 is moving toward persistent assistant behavior while trying to keep Proton’s core promise intact: saved context should not become server-readable user profiling. The announcement does not detail every safeguard around memory, so the key question is how clearly Proton exposes controls for reviewing, editing or deleting what Lumo remembers.
For readers tracking the split between hosted private AI and self-run model infrastructure, the trade-offs are different from the ones in One Command Spins Up a Private vLLM Server on HF Jobs. Proton is not asking users to operate their own inference stack; it is asking them to trust Proton’s encrypted service design.
Lumo’s web search upgrade also changes its usefulness. Earlier Lumo material from Proton said users could ask the assistant to search the web for new or recent information, while its default knowledge remained separate from live browsing. Lumo 2.0 now puts more weight on live results and citations, which should make source-checking more visible.
| Lumo feature | Lumo 2.0 change | Privacy claim attached |
|---|---|---|
| Images | Analyze, edit and generate images | Protected by zero-access encryption |
| Memory | User-controlled memory, encrypted Projects, Custom Lumos | Personalized workflows without Proton reading saved chats |
| Web search | Live results and source citations | Private search experience inside Lumo |
| Business use | Team access management | Conversations not logged or used for model training |
Benchmark gains give Proton a performance argument, not just a privacy one
Proton is also making a performance claim around Lumo 2.0 Max. The company says the model scored 240% higher than Lumo 1.4 on the Artificial Analysis Intelligence Index.
That index combines results from nine benchmarks, including GDPval-AA v2, Terminal-Bench v2.1, Humanity’s Last Exam, and GPQA Diamond. Proton also says Lumo 2.0 Lite scored 127% higher than Lumo 1.4 on the same index.
The benchmark claim gives Proton a second argument. Lumo is no longer being positioned only as the safer option; Proton wants to argue that its assistant is becoming competitive enough for everyday work.
Andy Yen, Proton’s founder and CEO, made that point directly in the launch comments quoted by 9to5Mac:
“Lumo 2.0 has been re-engineered from the ground up and the introduction of thinking mode gives it powerful new capabilities. User testing demonstrates that the gap has closed to the point that for many use cases, users can no longer perceive a qualitative difference between Lumo 2.0 Max and the latest models from OpenAI and Anthrophic.(sic) […] Lumo 2.0 demonstrates that users no longer need to choose between powerful AI capabilities and meaningful privacy protections.”
That is a strong claim, but it is still Proton’s claim. The announcement does not provide independent user testing data, task-level results, or side-by-side examples showing where Lumo 2.0 Max matches those rival models.
The same caution applies to “thinking mode.” Proton says it gives Lumo 2.0 new capabilities, but the available material does not explain how it works, when it activates, or what trade-offs it introduces in speed, cost or transparency.
For readers following model-release claims more broadly, Too Powerful for Public? Claude Fable 5 Hits Users is useful context on how capability narratives can run ahead of what users can verify in daily use.
Free access remains, but the practical limits sit in the paid tiers
Lumo 2.0 is available now. Proton says core AI features are included in the free tier for “everyday private use.”
There are also two paid plans. Lumo Plus includes unlimited chats, Projects, advanced image generation, and access to Proton’s most capable models. Lumo Professional is aimed at teams that need secure AI collaboration.
The source material does not list pricing. It also does not specify detailed usage caps for the free tier, limits on image generation, or the exact separation between Lumo Plus and Lumo Professional beyond the feature descriptions.
Those omissions matter because Lumo 2.0’s appeal depends on more than privacy promises. Users will judge whether the free tier is useful enough, whether paid limits feel restrictive, and whether the most capable models are available where they need them.
The next decision point is daily trust, not launch-day features
Proton’s bet is clear: enough users want AI tools, but do not want their prompts, files and business material logged, shared or used for training. Lumo 2.0 is built to turn that anxiety into product adoption.
The immediate test is whether Proton can make privacy feel invisible rather than limiting. Image generation, memory and live search all raise the stakes because they invite users to give the assistant more context.
If Lumo 2.0 can deliver reliable answers, useful image workflows and clear controls over memory, Proton has a sharper case for privacy-first AI. If those features lag the larger assistants, Lumo risks becoming a tool users trust with sensitive tasks but skip for everything else.
The Bottom Line
- Proton is trying to match mainstream AI assistant features without weakening its privacy-first pitch.
- Lumo 2.0 expands beyond chat into images, memory, projects and live web-connected answers.
- The launch strengthens Proton’s push into business AI for organizations concerned about data exposure.
