Proton Lumo Desktop: What IT Should Check Before Connecting Work Apps

Proton Lumo Desktop connects to work apps, reads local files, and runs scheduled tasks. See which permissions IT should test before granting access.

Written By
Michelle Lojo
Michelle Lojo
Nov 19, 2026
Proton Lumo Desktop: What IT Should Check Before Connecting Work Apps

Proton’s Lumo Desktop app can connect to local files and work apps and run tasks on a schedule. Image: Proton.

Connect a Proton Mail inbox to Lumo Desktop and, with the default permissions, the assistant can send or delete messages without asking first.

Proton launched Lumo Desktop on Nov. 17 for Windows, macOS, and Linux. The app requires a Proton account and gives the AI assistant access to local files and connected services. It can also run tasks on a schedule.

For IT teams, a weekly summary is an easy first test. The bigger decision is whether the setup that lets Lumo read an inbox should also let it send mail or delete messages.

What Lumo Desktop can access

In its launch announcement, Proton listed integrations for Google Drive, Jira, and Confluence. Proton Mail connects through Proton Mail Bridge, which requires a paid Proton Mail plan. Local tools let Lumo work with files, fetch browser content, and run shell commands. The app also supports local and remote servers that use the Model Context Protocol (MCP).

Advertisement

Proton’s connector setup guide says actions are set to Auto-approve by default. Users can change each action to Ask human or Deny. Connected tools are also enabled by default in chats, where users can turn individual connectors off through the Tools menu.

That makes scope especially important for teams that add their own integrations. MCP tool descriptions can influence an agent’s actions, so IT should review which servers and tools it exposes to Lumo before connecting them to work data.

Scheduled tasks have their own permissions

Lumo Schedules can run a task once or on a recurring basis. Each task has its own connector and permission settings. A help desk could use a schedule to summarize messages awaiting a reply, then use a Skill to give each summary the same format. Skills store reusable instructions and can include templates or reference files.

Proton says it keeps saved chat history under zero-access encryption and excludes conversations and data from connected apps and files from model training. Its security documentation says Lumo processes prompts on Proton’s servers. A local file connector gives the app a way to retrieve files from the device, so teams should review what information a workflow sends for processing before using it with work documents.

Advertisement

More must-read AI coverage

What IT should check before a rollout

Start with a test mailbox or a folder of approved sample files. Create a scheduled task with only the read actions it needs, then select Run now. Check that the summary uses only the intended messages or files before relying on the schedule.

  • Verify which actions need approval. In a test mailbox, allow Lumo to read messages, set sending to Ask human, and set deletion to Deny. Confirm that Lumo can summarize messages, pauses for approval before sending, and blocks deletion. Configure and test those settings separately for the scheduled task.
  • Limit the sources and tools. Keep shell access off for a mailbox or document-summary pilot. Use an account with access only to the test material. Email and documents can carry instructions that an assistant may mistake for part of its task, a risk known as indirect prompt injection.
  • Test how to stop and investigate a run. Pause the schedule, disconnect the test account, and verify what the workflow can still access. Check whether your team can tell which account and connector a run used and what it did. That evidence matters when deciding how to monitor AI agent activity across work systems.
Advertisement

A successful summary can show that Lumo saves time. Expand the pilot when IT can verify that the assistant stays within its approved scope and that the team can cut off access if a workflow misbehaves.

Read more: Find out which AI guardrails to test before an assistant can change records or take other actions in production.

Michelle Lojo

Michelle Lojo is the News Editor for TechRepublic, with eight years of experience in journalism. She oversees timely, accurate, and accessible coverage that helps readers understand the technologies, companies, and developments shaping the industry. Michelle is committed to turning complex topics into clear, relevant stories for technology professionals and business leaders.