Research for a more capable tomorrow

Modular intelligence for an open future

We research methods to make language models more composable, understandable, and reliably controllable — so new capabilities can be added, adapted, and shared with predictable effects.

Capability control

Add, adapt, and remove capabilities with confidence.

Model adaptation

More capable models with fewer trade-offs.

Real-world impact

From research to practical applications.

Each capability attaches to the same base model and comes off cleanly, without disturbing the others. This is an illustration, not a live model.

A more open, composable future

We share our work through open research, collaborations, and tools for the community.