Lumo 2.0 Adds Images and Memory While Keeping Chats Out of Training

Proton’s Lumo remains available, but it is substantially broader than the text-focused assistant released on July 23, 2025. The official Lumo update timeline dates Lumo 2.0 to June 30, 2026 and records a further data-visualization feature on August 3, 2026.

The central proposition has not changed: Lumo combines generative AI with a policy against using conversations for model training. What has changed is the range of information it can process and retain for the user, including images, web results, project files and selected context across conversations.

How Lumo changed after its original launch

The first release covered familiar chatbot tasks such as answering questions, drafting text, summarizing uploaded material and helping with code. Its main point of distinction was the handling of conversation data rather than an exclusive kind of AI output.

Lumo later gained Projects, which group files and instructions around a continuing task. Lumo 2.0 then added image recognition, generation and editing, improved web research with citations, customizable assistants and memory that can preserve selected preferences or working context between sessions.

A contemporaneous TechCrunch report documented the June 30 release, including image tools, a Thinking mode, user-controlled persistent memory in Projects and immediate access through free, Plus and Professional tiers. Its performance and privacy statements ultimately originated with Proton, so they should not be treated as results from an independent benchmark or security audit.

The product now offers Lite and Max model choices as well as Fast and Thinking modes. Those labels describe the intended balance between speed and more involved processing; they do not guarantee that a slower mode or larger model will produce a correct answer.

What Lumo’s privacy model actually covers

Lumo’s privacy proposition consists of several separate commitments. Proton’s current Lumo privacy documentation states that chats are not used to train AI models, processed prompt data is erased, conversation logs are not retained, and saved histories are synchronized with zero-access encryption.

Zero-access encryption applies to stored chat history. It is designed so that only the account holder can decrypt that history after signing in. It does not mean the AI can answer without processing a prompt: submitted material still has to reach the servers running the model so a response can be generated.

Guest, Free and Plus use also differ in how a conversation persists. Guest sessions are designed to disappear when the session ends, while signed-in users can keep searchable history on their devices and synchronize its encrypted form. Memory is a separate feature: it preserves selected context so Lumo can reuse it in later conversations.

These distinctions matter because “no logs,” encrypted history and persistent memory describe different parts of the system. A service can avoid retaining server-side prompt logs while still storing an encrypted history chosen by the user, and it can remember selected preferences without making the plaintext history readable to the provider.

The privacy architecture does not resolve whether a person is authorized to submit particular material. An employment contract, medical record, client file or another person’s correspondence may remain restricted even when the AI provider promises not to train on it. Encryption protects access to data; it does not supply consent or override workplace, contractual and legal obligations.

What the expanded assistant can handle

Lumo 2.0 is no longer limited to a text-in, text-out workflow. Its functions now span several common uses:

• drafting, editing, restructuring and translating text;
• summarizing documents and answering questions about uploaded files;
• reviewing code and explaining technical material;
• searching the live web and presenting cited results;
• analyzing, generating and editing images;
• organizing related material inside Projects;
• reusing instructions through Custom Lumos;
• turning information in documents and spreadsheets into visualizations.

Projects and memory are the most consequential additions for continuing work. A one-off chatbot forgets useful context at the end of an exchange; a persistent workspace can retain files, instructions and preferences. The same convenience makes it important to distinguish temporary conversation data from information intentionally kept for later use.

Web search also changes the character of an answer without making it automatically reliable. Access to current pages can reduce dependence on an older model knowledge cutoff, but the assistant may still select a weak page, misunderstand it or attach a citation that does not support the surrounding statement.

Image and visualization tools introduce similar limits. A generated picture can contain invented details, while a chart can misinterpret headings, units or the relationship between columns. The polished appearance of either output is not evidence that the underlying interpretation is accurate.

Free access and paid tiers

Lumo continues to offer a free entry point alongside paid access. The practical difference is not that the free version lacks the core chatbot entirely, but that model access, message volume, image generation, history and project capacity can be constrained.

Exact allowances and subscription prices are unsuitable as permanent reference points because plans can change. The durable distinction is between occasional access for evaluating ordinary tasks and paid tiers intended for heavier use, larger workloads or more sustained access to advanced models and features.

An account also changes the experience independently of payment. Signed-in use enables encrypted history across sessions, whereas guest use is oriented toward temporary conversations. Neither arrangement verifies the factual accuracy of an answer or grants permission to upload protected information.

Privacy does not answer the quality question

Lumo’s clearest difference from mainstream assistants remains its approach to retention, training and encrypted history. That can be material for people working with unpublished notes, personal documents or other information they do not want absorbed into a provider’s training pipeline.

It does not establish that Lumo is the strongest model for every research, coding or creative task. The breadth of Lumo 2.0 brings it closer to the expected feature set of modern assistants, but capability claims from the developer and independent comparisons answer different questions.

Generative AI can fabricate names, dates, quotations and references even when its interface presents them confidently. Search citations should be opened and checked against the exact claim, while document summaries should be compared with the relevant passages before they inform a consequential decision.

The meaningful change is that Lumo now pairs its original privacy commitments with multimodal tools, cited web research and persistent workspaces. For newcomers, the important distinction is equally clear: stronger controls over conversation data can reduce one category of risk, but they do not guarantee accurate output, lawful use of uploaded material or complete confidentiality outside the boundaries described by the service.

Also read:

• While Hiring Moving Company 9 Things to Look Out for
• Eminem Promoted BetterHelp While Its Privacy Refunds Were Underway
• YouTube Adds Conversation History to Ask Studio AI Chatbot