Proton has released desktop apps for its Lumo AI chatbot on all major desktop OSes, including Linux.
Lumo is the Swiss privacy company’s own AI assistant, introduced last year and available to use on line. It offers the same sort of features as other generative AI chatbots, including text and image generation, memory, web searching and editable canvases.
It runs on ‘open models’, touts better privacy safeguards and encrypts users’ chat history. It’s free to use (with limits) on a free Proton account. Users with paid plans can use additional features, and enjoy higher usage limits.
The GitHub repo for Proton’s Lumo Desktop says it “wraps the Lumo web app in a native Tauri shell and gives it access to your computer and your tools, with your approval”.
With the desktop app, Proton adds three new features to Lumo that allow its AI to do more than sit passively awaiting your input.
Connectors let you grant Lumo access to local folders and run commands, with controls over it can do access with or without your explicit approval. And you can connect to a variety of online services as well as any MCP server.
Schedules let you create tasks that run, well, on a schedule, e.g., connect it to your Proton Mail account and it can go through your inbox once a week to provide a summary of what’s there (accurate or otherwise).
Skills let you define files with instructions/prompts to re-use in a conversation, rather than continually copy and paste from a .txt file or one of your earlier conversations.
If you fancy checking it out – reminder: a free Proton account is required to use it – you can download Lumo Desktop for Windows, macOS and Linux (RPM and Debian installers are provided on the latter; supports Ubuntu 22.04 LTS or later).
How open Lumo’s models are is up for debate. On launch, the European Open Source AI Index described Lumo as the “least open ‘open’ AI assistant” in a blog post.
Proton CEO, Andy Yen, has also attracted criticism of late for his aggressive AI push within the company, referring to one employee’s request for time to perform proper security testing of AI generated code as ‘loser thinking‘.