EPFL · Science

Apertus AI Model Integrated into Proton's Lumo Assistant

Researchers behind the open-source AI model Apertus are partnering with Proton, a Swiss privacy-focused technology company, to integrate Apertus 1.5 into Proton's AI assistant, Lumo. This collaboration aims to provide Apertus with real-world user feedback for model improvement and foster European AI sovereignty.

The partnership allows Apertus researchers to receive feedback from millions of Lumo users, a crucial element typically lacking in academic open-source models. Users can opt-in to share anonymized feedback to aid in Apertus' development, while all conversations remain private and protected by zero-access encryption.

Apertus 1.5, developed by researchers at EPFL and ETH Zurich with partner institutions, is trained on the Alps supercomputer and released under an Apache 2.0 license. Its development process, training data, and source code are transparent, allowing for inspection and adaptation by researchers, public institutions, and companies.

The model is trained on publicly available data, respects website opt-outs, and is designed to remove personal data at scale. This initiative contributes to greater technological sovereignty in Switzerland and Europe by providing a competitive, locally developed AI alternative.

Since its launch in September 2025, Apertus has moved beyond research into practical applications. It is used in various settings, including on-premises installations and cloud services. Lumo's integration marks a significant step in bringing the model to a wider user base and feeding learnings back into the project.

AI-samenvatting op basis van de bron.