Skip to main content

Proton Lumo Desktop: Connectors, Schedules and What Stays Private

Proton launched Lumo Desktop for Windows, macOS and Linux on 17 November 2026. It adds Connectors (local files, shell, Proton Mail, Google Drive and MCP servers), Schedules for recurring tasks, and Skills. Chat history uses zero-access encryption, but prompts are still decrypted on Proton's European servers to run the model, so this is private cloud AI, not on-device AI.

On this page

What Proton released on 17 November

Proton, the Swiss company behind Proton Mail and Proton VPN, released Lumo Desktop on 17 November 2026, its first desktop app for the Lumo AI assistant, with builds for Windows, macOS and Linux. Alongside the apps it introduced three features: Connectors, Schedules and Skills. Skills are reusable instructions that carry across conversations, and Proton says existing Custom Lumos will be migrated to Skills automatically.

The more important change is what the desktop app can reach. Until now Lumo was a chat window. Proton's FAQ says Connectors and Schedules are, for now, only available in the desktop app, which makes Lumo Desktop the company's first step from a private chatbot towards a private agent that acts on your machine.

Connectors give Lumo access to your files and apps

Connectors let Lumo find information and take actions in the tools you link to it. Proton lists local tools such as filesystem access, shell execution and fetching browser content, Proton Mail through Proton Mail Bridge, third party services such as Google Drive, and your own local or remote MCP servers. For every connector you can set each action to run automatically, require approval, or be blocked.

Read the defaults before you connect anything sensitive. Proton's support page says connector actions are set to auto-approve by default, and connected tools are enabled in conversations by default. For a connector that can run shell commands, switching risky actions to "needs approval" is a sensible first step.

Schedules let Lumo run a task at a set time or on a recurring basis, and each scheduled task has its own permissions. Proton's support page does not say whether the desktop app has to stay open for a scheduled task to run.

Where your prompts are actually processed

Proton's privacy pitch is strong: it says Lumo never uses your conversations, or information from connected apps and files, to train models or serve personalised ads, and that zero-access encryption means no one but you can read your chat history, not even Proton. Proton also says Lumo keeps no records of when you chatted or what you discussed.

It is important to be precise about what that covers. Proton's own security page says messages are encrypted in transit, and that the server decrypts each message temporarily to run it through the language model. Zero-access encryption protects your stored history; it does not mean the model runs on your computer. Lumo runs on servers Proton controls in Europe, and the launch post says the service is protected by Swiss privacy law and sits outside the reach of the US CLOUD Act.

So Lumo Desktop is local in its reach but not in its inference. A file a connector reads on your laptop is sent to Proton's servers to be processed, under Proton's no-logging and no-training promises.

Who Lumo Desktop suits

If you want an assistant that can work across your files and inbox, and you trust Proton's jurisdiction and policies more than a US cloud provider, Lumo Desktop is one of the more privacy-minded options. Proton offers a free tier with limited usage and a Lumo Plus plan for higher limits, and says it is publishing the source code for the new tools so the community can inspect them.

If your rule is that private data never leaves the device, the trade-off is unchanged: you still need a model running locally. The desktop app also widens what an assistant can touch, so the same habits apply as with any agent: connect only what you need, keep shell access behind approval, and review scheduled tasks regularly.

CuriousLM runs supported AI models locally on your device. Try CuriousLM.