We envision a world where AI can take part in everyday life. People can think aloud, develop ideas together, and get things done with intelligence that understands the situation. As models become more capable, people should be able to discover what becomes possible without having to specify everything in advance.

Interact is a research lab building the models that make this kind of participation possible. We’re developing a general interaction model that works with the most capable reasoning and execution models available. Our goal is to build the interface between humans and superintelligence.

Our core focus is understanding the context and intentions behind an interaction: when to listen, when to contribute, and how to keep work aligned as people change their minds. We are starting with speech and plan to expand into vision, toward an always-on model people find useful and reliable enough to keep with them.

Our team brings together researchers and engineers from Stanford, Berkeley, MIT, and Cambridge.