Proton has expanded its Lumo AI assistant with connections to personal services, recurring scheduled tasks and desktop apps for Windows, macOS and Linux. The key change is practical: people can give the assistant access to selected information while setting limits on what it may do.
Watch Desk analysis
What happened
Digital Trends reports that Lumo’s new Connectors can find information in Proton Mail, Google Drive, Jira and Confluence, as well as local tools and custom Model Context Protocol servers. Users can set different permissions: for example, allow Lumo to read emails, require confirmation before it sends a reply, and prevent it from deleting anything.
Schedules let users automate recurring requests, such as preparing a weekly email roundup. Each scheduled task has its own permissions and can be modified, paused or removed. Skills are reusable instructions for preferences such as email tone or report formatting. The report says Proton has also made Lumo’s source code public and claims that connected-service information and conversations are not used to train its models or deliver targeted ads.
Why it matters
An assistant that can search across services or act on a schedule is more useful than one that only answers questions, but it also needs access to more personal information. Lumo’s task-by-task controls offer a concrete way to limit that access and review what recurring jobs are allowed to do.
Our read
This is a meaningful step towards AI assistants that can do useful work across apps without asking users to hand over the keys to everything. The permission controls are the part to pay attention to, not just the longer feature list. Start with narrow access and a low-stakes task; convenience is lovely, but deleting the wrong thing is a remarkably efficient way to lose it.
What to watch
- Which connectors and local tools are supported in practice.
- How clearly permissions apply to scheduled tasks and connected services.
- Whether Proton’s privacy claims are backed by enough technical detail for users to assess them.
Discussion spark: Would you let an AI assistant read your email or files if you could require approval before it sends or changes anything, or is connecting personal accounts a line you would not cross?
Sources and evidence
- Source update (18 November 2026, 08:40 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.