Proton has released Lumo 2.0, an update to its privacy-focused AI assistant that adds reasoning, image generation, stronger web search and a memory feature called Projects. Proton says Lumo keeps no server-side conversation logs, does not use chats to train models, and stores chat history with zero-access encryption...
Proton has released Lumo 2.0, a major update to its privacy-focused AI assistant with new reasoning, image and memory features.
In a company blog post titled “Lumo 2.0: The most powerful private AI,” Proton says the updated assistant adds reasoning capabilities, image generation, stronger web search, a memory feature called Projects and code that users can inspect. Proton positions Lumo as an AI assistant designed around privacy controls rather than ad targeting or data collection.
TechCrunch also reported that Proton’s Lumo upgrade includes image recognition and generation, faster responses, a “thinking mode,” Projects memory, zero-access encryption, no server-side logs and no customer-data training.
The update builds on Proton’s earlier Lumo launch. In its introductory post, “Introducing Lumo: AI where every conversation is confidential,” Proton said Lumo does not use conversations to train AI models and does not keep server-side conversation logs. The company also said saved chats use zero-access encryption, meaning Proton says it cannot read stored conversation history.
Proton’s separate post, “Lumo security model: How Proton makes AI private,” gives a more detailed explanation of how the system works. According to Proton, messages are encrypted as they pass through Proton’s internal systems before reaching the LLM server. The company says stored conversation history is protected with zero-access encryption, so only the user’s device can decrypt it.
That security model is important because “zero-access” does not mean every step of AI processing happens while data remains unreadable. Proton’s security-model post says messages must be processed by the LLM server, which means prompts are decrypted for model inference at that stage. Proton’s privacy claim is therefore narrower and more specific: it says it avoids retaining server-side logs, does not train on user conversations, and protects stored chat history so Proton cannot read it later.
The most visible additions in Lumo 2.0 are aimed at making the assistant more competitive with general-purpose AI chatbots. Proton says reasoning support is intended for more complex tasks, while improved web search is meant to help the assistant retrieve and use current information. Image generation and image recognition expand Lumo beyond text-only interactions.
The Projects feature adds a form of memory. According to Proton and TechCrunch, Projects allows Lumo to keep context for ongoing work. Proton frames this as a privacy-preserving memory system, while continuing to state that customer conversations are not used for model training.
Proton also says Lumo 2.0 includes code inspection options. The company’s blog describes this as part of its transparency approach, though the available source material does not establish that every part of Lumo’s infrastructure or model stack has undergone an independent third-party audit.
Proton’s claims are detailed and consistent across its own launch, update and security-model posts, and TechCrunch corroborates the main product changes. However, the strongest privacy and encryption claims still depend largely on Proton’s own documentation.
For users who choose Lumo because of privacy, the distinction matters. Proton says it does not log conversations on the server side, does not train AI models on user conversations, and encrypts stored chat history so Proton cannot read it. Its own security model also makes clear that the AI system must process decrypted prompts at inference time.
An independent third-party audit of Lumo’s encryption and data-handling implementation would give users more evidence beyond company statements. Based on the cited sources, Proton has published a security explanation and product details, but the provided source material does not show that such an independent audit has been completed for Lumo 2.0.
Lumo 2.0 is Proton’s attempt to make a privacy-focused AI assistant more capable without abandoning its stated limits on logging and training. The update adds features users expect from modern AI assistants, including reasoning, web search, image tools and memory. The central question is not only what Lumo can do, but how much users can independently verify about its privacy guarantees.
Proton has released Lumo 2.0, a major update to its privacy focused AI assistant with new reasoning, image and memory features.
Proton positions Lumo as an AI assistant designed around privacy controls rather than ad targeting or data collection.
In its introductory post, “Introducing Lumo: AI where every conversation is confidential,” Proton said Lumo does not use conversations to train AI models and does not keep server side conversation logs.
Continue reading
