Who we are
We are an AI lab founded to make open-source models usable by everyone.
Why we exist
Every week a new open-source model is released in some field of science: models that predict protein structures, map flood zones from satellite images or transcribe lecture recordings. But before a researcher can use them, they need a powerful GPU and days of setup work.
We take on that burden. We install the models, test them and deliver them to researchers ready to use.
How we work
- Selection: We review models that are useful in their field, along with their licenses.
- Installation: We install each model in its own environment and record its GPU requirements and setup steps.
- Testing: We use fixed test inputs for each category, so models can be compared with one another.
- Report: We share sample outputs, speed and errors encountered on each model's page.
- Access: Models that pass testing are opened to approved members.
Infrastructure approach
Models are not kept running all the time. When a request comes in, the model starts on a GPU and shuts down when the job is done. This lets us offer a large number of models at a reasonable cost.
Our transparency principles
- We never show a model as "ready" unless it is running.
- We list every model's license on its page and clearly mark models that are closed to commercial use.
- Health and psychology models are for education and research only. They are not diagnostic tools.
- The cybersecurity category includes defensive models only.
For universities
We can build a set of models around your departments' needs and open it to your campus: medical models for health departments, protein and chemistry models for pharmacy, image, 3D and music models for fine arts, and more.