Lumo

Lumo is Proton’s answer to ChatGPT, and it arrives with the one thing most privacy-branded assistants cannot show: a stack you can inspect. The apps are open source, and so are the models behind them. Proton runs those models on its own servers under Swiss law rather than routing your prompts to a US provider, which is the distinction that actually matters if your reason for looking is where your data ends up.

The privacy claim is stronger than the usual "we do not sell your data". Conversations use zero-access encryption, so Proton cannot read your chat history even if it wanted to, and prompts are not used to train anything. You can reach it over Tor, and guest access lets you use it without an account at all, which is a genuinely unusual thing to offer in this category.

What Lumo is not is a frontier model. It runs open-weight models that Proton has optimised rather than something trained in-house, so on hard reasoning it sits behind the American assistants and behind Mistral. That is the trade being offered, and Proton is reasonably open about it.

What runs underneath

Lumo is powered exclusively by open-weight models, currently Qwen 3.5 and GLM 5.2 for text, with Image-Turbo and FireRed-Image-Edit-1.1 for images. Proton optimises and hosts them itself, which is what makes the no-logging promise enforceable: there is no third-party API call that could retain a prompt.

Worth knowing, and rarely mentioned: those base models were trained by Chinese labs, not European ones. The weights are open and the inference happens in Switzerland, so no data leaves Proton’s infrastructure, but if your definition of European sovereignty extends to who trained the model rather than who runs it, Le Chat is the stricter answer. If it extends to where your conversations are processed and stored, Lumo is.

The privacy architecture

Zero-access encryption means your chat history is encrypted with keys Proton does not hold. That is the same model as Proton Mail and it has the same consequence: lose your password and recovery phrase and the history is gone, because nobody can decrypt it for you.

Guest access and Tor support cover the case where even holding an account is more exposure than you want. Neither is a gimmick here; both follow from the same design as the rest of Proton’s products.

Where it fits in the Proton suite

Lumo reads files from Proton Drive, which is the practical reason to pick it over a standalone assistant if you are already inside the suite. Projects and custom assistants keep context across conversations, and Lumo for Business is sold through Proton Workspace rather than as a separate contract.

Proton Mail, Proton Drive, Proton Pass and Proton VPN are all in this directory already. Lumo is the fifth product and the one that puts an assistant inside a stack a lot of privacy-minded European teams have already standardised on.

Limitations to Consider

On difficult reasoning, long-context work and coding, Lumo is behind ChatGPT, Claude and Gemini, because it runs open-weight models rather than a frontier model of its own. There is no self-hosting option: you trust Proton’s servers or you use something else. The free tier is tight enough that regular use effectively means Plus, and the underlying base models come from Chinese labs even though the inference is Swiss.

Who Should Use Lumo

Anyone whose objection to the American assistants is about jurisdiction and data retention rather than raw capability, and especially teams already using Proton Mail and Drive. If you need frontier-level reasoning, this is not yet that; if you need to be certain your prompts are not retained or trained on, few alternatives can show their working this clearly.