No install
Contribute from a browser tab. Your device helps train the model while it sits idle.
Grounded in research
Built on a decade of work in federated and communication-efficient learning.
Community-owned
The resulting model belongs to its contributors, not a single company.
What is this?
Lupus is an early-stage research project exploring whether a useful foundation model can be trained collaboratively, across thousands of ordinary computers, instead of inside a single data center. The idea draws on a well-established field of machine-learning research called federated learning — and on recent breakthroughs that make training large models across slow, unreliable connections finally practical.
The detailed methods are still being developed and are intentionally kept brief here. The article below explains the field and our direction in plain language, along with a high-level view of how the system is designed to work.