
by Proton
Proton's private AI chat app: no logs of what you ask, zero-access encrypted history, and a guest mode whose chats are erased at the end of each session.
Pick Lumo when what you type is private: a health detail, a family matter, a CV gap. It replaces a mainstream chatbot for those chats, not for hard coding or research, where an App Store reviewer got outdated answers until web search was on. Its 580 App Store ratings average 3.8.
A published, plainly worded encryption design with no logs and a guest mode that leaves nothing behind.
The trust built by Proton Mail and VPN and the Proton Account that ties them together give it a loyal base, but the models are open-weight ones anyone can serve.
Main trade-off
Answer quality: a 4-star App Store review describes outdated replies until web search is enabled.
| Plan | Monthly price | What it includes |
|---|---|---|
| Free | Free | Limited messages and chat history, Limited Lumo 2.0 Max use, Limited image generations, 1 project, No credit card |
| Lumo Plus | $12.99/mo | More Lumo 2.0 Max use, More messages and chat history, More image generations, Unlimited projects, Unlimited custom Lumos, $9.99 a month when billed yearly ($119.88) |
Read on proton.me/lumo on 17 November 2026: Free, $0 with no card (limited messages and history, limited Max model use, 1 project). Lumo Plus is $12.99 billed monthly, or $9.99 a month billed $119.88 a year, with a 30-day money-back guarantee. Proton also shows CHF and EUR prices, and sells separate Lumo for Business plans.
Lumo is a privacy-first AI chatbot from Proton. It runs open-weight models such as Qwen 3.5 and the Swiss Apertus 1.5, keeps no logs of prompts or replies, and stores saved history with zero-access encryption. A guest mode lets anyone chat without an account, and Apertus was trained on over 1,000 languages.
No-logs chat
Prompts and replies are not logged on Proton servers, and the model servers hold no memory between sessions.
Zero-access encrypted history
Saved chats on Free and Plus are encrypted so only you can read them after signing in to your Proton Account.
Guest mode
You can chat without an account, and a guest conversation is erased at the end of the session.
Choice of open-weight models
Lumo 2.0 Lite for everyday tasks, Lumo 2.0 Max for harder ones, and Apertus 1.5, a Swiss open model trained on over 1,000 languages.
Opt-in web search
Search stays off until you switch it on, so a prompt does not leave for the web unless you choose.
Projects and image generation
Projects group chats and files, with one project on Free and unlimited on Plus, and Lumo can generate images within plan limits.
| Tool | Price from | Free plan | Runs on | Rating |
|---|---|---|---|---|
| $12.99/mo | Yes | Web · iOS · Android | 3.6 | |
| $8/mo | Yes | Web · iOS · Android · Mac · Windows | 4.2 | |
| $20/mo | Yes | Web · iOS · Android · Mac · Windows | 4.3 | |
| $18/mo | Yes | Web · iOS · Android | 3.8 |
Open-source app code; Proton service
On 17 November 2026 Lumo Plus was $12.99 billed monthly, or $9.99 a month when paid yearly at $119.88, with a 30-day money-back guarantee. The free plan costs nothing. Business plans are sold separately.
Yes. There is a free plan with no card on file that gives limited messages, history and image generations, with a daily cap on the stronger Max model. You can also chat as a guest with no account, and that chat is erased when the session ends.
No, by Proton's own account: its privacy page says chats are not kept or used for model training. The only route into training is feedback you opt to send on Apertus answers, which is anonymised before it reaches Proton and the Swiss university teams that build that model.
Not in the usual sense, and Proton says so. Your prompt is encrypted on your device with a key only Lumo's servers can open, because the model must read it to answer, and Proton calls this user-to-Lumo encryption. Saved history is stored with zero-access encryption that Proton cannot read.
Proton runs open-weight models on its own servers rather than one in-house model. Its privacy support page currently lists a Qwen release, a GLM release and the Swiss Apertus model, shown in the app as Lite, Max and Apertus. The line-up changes, so check the models page for the current set.
Lumo is the AI chat assistant from Proton, the Europe-based maker of Proton Mail and Proton VPN, built by the team behind those products and the nonprofit Proton Foundation. It sits in Hokai's AI chatbots list, and its whole pitch is privacy: Proton says it keeps no logs of what you ask or what Lumo answers, never uses chats to train models, and stores saved history with zero-access encryption that Proton itself cannot read.
Under the hood it runs open-weight models rather than one in-house model. Proton's privacy page names Qwen 3.5, GLM 5.1 and Apertus 1.5, the last built by EPFL, ETH Zurich and the Swiss National Supercomputing Centre, and its models page groups them as Lumo 2.0 Lite, Lumo 2.0 Max and Apertus. Prompts are encrypted on your device with a per-request key that only Lumo's GPU servers can open. Proton is careful to call this user-to-Lumo encryption rather than end-to-end encryption, because the model has to read your message to answer it. Web search is off until you switch it on.
It suits anyone who wants to think through something sensitive with an assistant: a health question, a family situation, a CV gap they would rather not explain to a data broker, or confidential work notes. You can try it as a guest with no account, and the apps run on iPhone, Android and the web.
The cost of that privacy is answer quality and features. One App Store reviewer, read in November 2026, rated it 4 out of 5 as good for privacy but not always the best for answers, after a chat gave outdated information until web search was turned on. Our own guest chat this month showed a different weakness: it wrote a clean CV line with the real dates, then padded the interview answer with caring duties we had not described. If you want the strongest model or a wide plugin and file ecosystem, ChatGPT, Claude and Gemini are more capable, provided you change their training settings. Venice is another privacy-first chat service with a larger image allowance, and fully local apps such as Jan or Ollama keep everything on your own machine at the cost of setup. Our private AI guide compares them by what each policy actually says.
Product facts on this page (pricing, plans, integrations, compliance) are read from Proton's own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.