# Lumo Desktop Setup: Check Connector Permissions Before Your First Automation

Proton's new desktop AI adds Connectors, Schedules and Skills. Here's where to start, what auto-approval means and what private processing does not mean.

By Paras Madan | 2026-11-18


Proton [launched Lumo Desktop on November 17](https://proton.me/blog/lumo-desktop-features) for Windows, macOS and Linux, alongside Connectors, Schedules and Skills. The practical change is that Lumo can work with connected information and tools rather than only the text you paste into a conversation.

Before connecting an inbox or scheduling work, I would review its action permissions. A privacy-focused assistant still needs clear limits on what it can read, send or change.

## Start with the official desktop download

Use [Proton's Lumo download page](https://proton.me/lumo/download), choose your operating system and sign in with your Proton account. Proton says you can get started with a free account; paid plans add features and higher usage limits. Connectors and Schedules are currently desktop capabilities, so do not expect the mobile app to offer the same setup.

I have not installed or tested these apps. The documentation reviewed does not establish a reliable minimum desktop OS version, so check the actual download requirements for your machine rather than assuming compatibility from the platform name alone.

## Review permissions before connecting tools

The [Connectors help page](https://proton.me/support/lumo-connectors) describes the setup path as Customize, then Connectors. Select a service, click Connect and complete its authentication. Custom MCP servers can use local stdio or remote HTTP connections.

The important default is easy to miss: Proton says connector actions begin set to Auto-approve. That means an action can run without asking you first. In a connector's Functions tab, you can change individual actions to Ask human or Deny. Connected tools are also enabled in conversations by default; use Tools, then Connectors to turn off those you do not need.

For a first trial, I would allow only the reading needed for one task, require approval for sending or modifying anything, and block deletion. That is a proposed starting configuration, not a claim about the exact actions every connector exposes. Read the actual function list before giving it access.

## Separate chat privacy from tool access

Proton's [security explanation](https://proton.me/lumo/security) distinguishes encrypted storage from processing. Messages travel encrypted to Proton's servers, where they are temporarily decrypted for the language model to generate a response. Saved conversation history uses zero-access encryption. Proton also says conversations are not used to train its models.

That is not the same as running the model entirely on your laptop. Nor does encrypted chat storage make a broad tool permission harmless. If a connector can send a message or modify a file, that action has consequences in the connected service. Decide whether the task needs that capability independently of whether you trust the chat-history protections.

## Test a narrow schedule before relying on it

The [Schedules guide](https://proton.me/support/lumo-schedules) uses Scheduled, then New task in the sidebar. You can describe the task to Lumo or configure it manually, including its instructions, connectors, frequency and permissions. Review those details before saving, then use Run now to inspect a first result. Tasks can later be edited, paused or deleted.

My proposed first task would be a read-only weekly document summary, not an automatically sent email. Specify the folder or source, time range and output format, and explicitly exclude file changes. Compare its answer with a few documents you already know. Missing a recent update is easier to catch in a small test than in a broad instruction to “keep me informed.”

The reviewed guidance does not clearly establish what happens when your computer is asleep or Lumo Desktop is closed. Do not make an important deadline depend on unattended execution until you have checked that behavior in your own setup.

## Add reusable instructions after the task works

Skills store instructions and can include templates or reference files. Proton says relevant Skills are applied automatically, and existing Custom Lumos will migrate to Skills.

I would first settle the summary format manually, then save that preference as a Skill. Keep it separate from action permissions: a writing-style instruction should not be treated as permission to send the writing. The useful first result is a repeatable task with understandable access, not the largest possible collection of connected apps.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical teams explain AI products and workflows clearly enough for people to use them.

[Book a 30-minute call](https://cal.com/paras-madan/30min).
