Proton has officially released Lumo Desktop for Windows, macOS, and Linux, bringing the internal AI tools used by their engineering teams to the general public. The release isn't just a chat interface; it includes three core features designed for practical workflow automation: Connectors, Schedules, and Skills. Crucially, Proton is making the source code for these tools publicly available, allowing developers to inspect exactly how the software handles data and executes tasks.

Connectors and Local Control

The Connectors feature allows Lumo to interact with local filesystems, shell execution, and browser content, as well as external services like Proton Mail, Google Drive, Jira, and Confluence. For developers, the standout capability is support for local or remote Model Context Protocol (MCP) servers, enabling custom integrations without relying solely on Proton's native connectors. Security is granular: users define which actions are automatic, which require manual approval, and which are blocked entirely. For instance, you can allow Lumo to search your inbox but block it from deleting messages.

Schedules and Reusable Skills

Lumo introduces Schedules to run tasks at specific times or on recurring intervals, such as weekly inbox summaries or monthly spending reviews. These tasks operate under their own permission sets, separate from the main application. Complementing this are Skills, which are reusable instruction sets that teach Lumo your writing style, career goals, or dietary preferences. If you have existing Custom Lumos, the system will automatically migrate them into the new Skills format, ensuring continuity for current users while standardizing how personalization is handled across conversations.

Sovereign Infrastructure and Privacy

Underpinning these tools is a significant infrastructure update. Proton has activated its first NVIDIA B300 cluster in Switzerland, built on Open Compute Project (OCP) standards. This move emphasizes sovereign AI infrastructure, keeping data processing within Swiss jurisdiction and outside the reach of the US CLOUD Act. The company reiterates that Lumo never uses user conversations or connected file data to train models or serve personalized ads, maintaining zero-access encryption for chat history.

Key Takeaways

  • Lumo Desktop is now available on all major platforms with fully inspectable source code.
  • Native support for MCP servers allows for deep, custom developer integrations.
  • Granular permission controls let users decide exactly what actions AI can take automatically.
  • New Swiss-based NVIDIA B300 infrastructure reinforces data sovereignty and privacy compliance.

The Bottom Line

Proton is finally giving us the private AI stack they've been building for themselves. By opening up the source code and supporting MCP, they're inviting the dev community to audit and extend their privacy-first workflow tools.