Lumo is Proton’s no-logs AI assistant

Lumo is Proton’s privacy-first AI assistant, launched on July 23, 2025, and its defining pitch is simple: Proton says it does not log chats, does not use them to train AI models, and protects saved history with zero-access encryption. That makes it a direct answer to a problem many mainstream chatbots still handle awkwardly: your prompts are often useful product data unless the vendor explicitly says otherwise.

Proton is the Swiss company best known for Proton Mail, Proton Drive, Proton VPN, Proton Calendar, and Proton Pass. In that lineup, Lumo is the company’s AI layer: a chatbot for writing, summarizing, brainstorming, file analysis, and search, built around the same privacy branding that already sells its encrypted communications and storage tools.

Lumo is Proton’s privacy-first AI assistant

At launch, Lumo was available on the web to both guests and logged-in users, with features that included question answering, drafting, summarization, translation, file uploads, and optional web search. Proton also said Lumo could work with files from Proton Drive, which is the obvious product fit: if your documents already live inside Proton’s ecosystem, the assistant is meant to help you use them without handing that workflow to a separate AI vendor.

That product positioning matters because privacy is where AI chatbots tend to get slippery. If a chatbot stores prompts, uses them for model improvement, or lets admins inspect usage by default, the “assistant” is also a data collection surface. That is why settings and architecture matter more than marketing copy, both for consumer tools and for enterprise chatbot privacy settings.

According to Proton’s getting started guide, guest users can try Lumo without creating an account. Logged-in users on Free and Plus plans get persistent features such as saved encrypted history, while Proton also offers Ghost mode, which lets signed-in users chat without saving a conversation.

Lumo is also not one single fixed model. Proton’s current support documentation says it is powered by open-source language models, and its model page currently describes Lumo 2.0 Lite and Lumo 2.0 Max, along with Fast and Thinking modes. That is a useful detail because it means “Lumo” is a product layer and privacy architecture, not one immutable base model in the way people often talk about ChatGPT or Claude.

Lumo’s privacy model centers on no logs and zero-access encrypted history

The strongest claim Proton makes is in its privacy documentation: Lumo chats are not logged by Proton and are not used to train AI models. Proton also says it does not share user conversations with third parties, and that Lumo operates under European data protection laws.

For saved conversations, the company’s model is more specific than a generic “we encrypt data” claim. Proton says chat history saved by signed-in users is protected with zero-access encryption, meaning the encryption keys are designed so Proton cannot read that stored content in plaintext. In practice, that puts saved history closer to Proton Drive or Proton Mail’s storage model than the usual cloud chatbot pattern, where the provider can generally inspect stored chats on its own systems.

“Your conversations are not used to train AI models, and Proton cannot read your saved chats.”

That does not mean the entire system is end-to-end encrypted in the strictest possible sense while a model is actively generating tokens. In its security model post, Proton says full end-to-end encryption for live AI inference is technically challenging, so Lumo uses transport encryption and asymmetric encryption to Proton-controlled GPU servers rather than full E2EE during model processing. That is a real limitation, stated plainly by Proton itself, and it is the tradeoff you hit when you want server-side AI to actually run.

Still, Proton’s setup is materially different from the default consumer chatbot pattern. Its support page says guest chats are erased when the session ends, while saved history exists only if a logged-in user chooses the account-based experience. That split matters. “No account” and “no saved history” is a cleaner privacy story than the more common setup where every interaction quietly accumulates in a permanent activity log. It also sits between two other poles: cloud AI services with broad retention, and local LLM privacy tradeoffs, where the privacy upside is stronger but the convenience and capability often drop.

The web search feature adds another boundary line. Proton says web search is optional and off by default, and its support materials note that enabling it lets Lumo pull in newer information beyond the model’s built-in knowledge cutoff. That is useful, but it also means the chatbot is no longer relying only on its base model and your local prompt context; it is reaching outward to fetch fresh information.

Lumo’s limits come from how AI processing works and which features are optional

On capability, Lumo looks like a fairly standard general-purpose assistant with a stronger privacy wrapper. Proton says it can summarize text, draft content, translate, answer questions, analyze uploaded files, and search the web when asked. For many users, that is enough. The differentiator is not some benchmark miracle; it is the promise that your chat history is treated less like free training fuel.

That promise lands because chatbot privacy is not an abstract concern. The stakes are obvious in everything from personal oversharing to workplace data leakage, which is why warnings about AI chatbot privacy risks keep recurring. A private assistant is only private if the vendor’s storage, retention, and training practices hold up under inspection.

There are also feature boundaries tied to plan level. Proton’s getting started documentation says Free and Plus accounts unlock different usage levels, and current support pages describe additional model choices such as Lumo 2.0 Max. Proton’s later product updates hub shows the service did not freeze at launch; it later added milestones including Lumo 1.3 and Lumo for Business. Independent reporting from TechCrunch also noted that Proton upgraded Lumo on June 30, 2026, confirming that the product has continued to evolve beyond its first release.

The cleanest way to describe Lumo, then, is this: it is a cloud AI assistant designed to minimize what Proton can keep, read, and reuse, not a magic system that removes every trust boundary from server-side AI. If you want absolute control, a local model still changes the equation more radically. If you want hosted AI with unusually strong privacy defaults, Lumo is one of the clearer attempts to build that.

The next concrete milestone is whatever Proton adds through its Lumo product updates page, which the company is using as the running log for new models, business features, and product changes.