Lumo by Proton: An AI Assistant That Doesn’t Spy on You (Finally)

I’ve lost track of how many “private” AI assistants I’ve tested that turned out to be just another way to funnel your data into some corporation’s training pipeline. So when Proton – the people behind Proton Mail and Proton Drive – announced Lumo, my skepticism was on full alert. I’ve seen too many tech companies use the word privacy like a discount sticker: slapped on for show, hiding a very different price underneath.

After a few weeks of reading, prodding, and checking Proton’s own documentation against what independent reviewers are finding, I’m prepared to say this: Lumo might actually be the rare AI tool that means it when it says it’s private. Not perfect, not feature-stuffed, but built for people who’d rather keep their conversations between themselves and their assistant – not themselves, their assistant, and 300 anonymous data annotators.

What Lumo Can Actually Do

On paper, Lumo handles the same kind of stuff ChatGPT or Claude will do for you:

• Summarising documents
• Drafting emails
• Writing or explaining code
• Answering questions and helping plan projects
• Analysing files you upload

If you’re in Proton’s ecosystem, you also get Proton Drive integration, which means you can pull in files from your encrypted storage straight into the chat. No awkward downloading and re-uploading. And yes, you can turn on web search – but only if you want to. By default, it stays off, which is already more than I can say for most of its competitors.

It works on web, iOS, and Android. If you want to try it without an account, you can. You’ll get a limited number of questions per week, and nothing will be saved. Log in with a free Proton account and you get 100 encrypted conversations per week. Go for Lumo Plus (about $12.99/month) and those limits vanish, plus you get bigger file uploads and an unlimited favourites list.

Privacy: More Than a Buzzword Here

Proton has built its reputation on encryption, so they’ve carried that over to Lumo. Here’s the chain of custody for your data:

1. Your prompt is encrypted locally.
2. That’s sent over TLS to Proton’s servers in Europe – specifically Germany and Norway.
3. The model processes it, generates a reply, then forgets it.
4. If you save your chat, it’s stored with zero-access encryption, meaning even Proton can’t read it.

They’ve also included Ghost Mode, which is exactly what it sounds like: when you close the chat, it’s gone. Poof. No lingering “history” quietly feeding back into training datasets.

Could they be lying? In theory, yes. Any company could. But Proton has a track record – they’ve fought legal battles over user privacy and published transparency reports for years. Combine that with their open-source model approach and public documentation, and you at least have a framework you can verify.

The Openness Angle

Proton says Lumo uses open-source models and that the code is available for audit. They haven’t published a permanent model lineup – it’s likely to change – but early reports suggest a mix of models from Mistral and NVIDIA’s OpenHands for certain tasks. That’s not unusual; even the big AI names swap back-end models quietly. The key difference here is that Proton isn’t pretending these are homegrown or keeping them entirely black-boxed.

What It’s Not

If you’re looking for an AI that will also be your voice assistant, run home automation, and book your next dentist appointment, Lumo isn’t it. This is not a “plug me into every part of your digital life” kind of tool. No plugins, no massive third-party integration library. And honestly, that’s a feature for anyone who sees “integration” as another word for “more ways to leak data.”

It also won’t match the sheer range of ChatGPT Plus with GPT-4o, Gemini Advanced, or Claude Pro. Those tools are stuffed with multi-modal tricks, huge knowledge cut-offs, and all the corporate polish in the world. Lumo is narrower. That’s the trade-off: you get less feature creep, more peace of mind.

My Real-World Take

I’ve been around enough tech launches to know that privacy branding often masks the same extractive logic as the rest of the industry. Lumo doesn’t feel like that. It feels more like a tool built for the Proton audience – the people who already use encrypted email, secure cloud storage, and don’t want to keep explaining to their friends why WhatsApp isn’t as “private” as it claims.

The catch is, you have to want exactly what it offers. It’s not going to out-flash the latest AI model from OpenAI or Google. It’s not going to run your smart home. What it will do is keep your chats yours, make them useful, and do it without asking for your trust – because the design means it doesn’t need it.

Quick FAQ

• Can I use Lumo without an account?
  • Yes, but you’ll be limited to a small number of weekly questions and nothing will be saved.
• Does it log my prompts?
  • No, Proton says your prompts are encrypted in transit and not stored unless you choose to save them.
• Where does my data go?
  • Proton-controlled servers in Europe, with zero-access encryption if stored.
• Is it as powerful as ChatGPT or Claude?
  • Not in raw range of features, but that’s the point – it trades scope for privacy.

If you want a fully private, encrypted AI assistant today, Lumo is at the top of a very short list. Just don’t expect it to solve every problem under the sun – expect it to solve the ones you actually want to keep between you and your machine.