Purpose Intelligence (Pi) is a conversational research prototype exploring an external control layer around existing language models to help AI assistants answer, ask, or wait.

Early experiments used small, controlled situations to inspect when a system proceeds, asks for clarification, or holds. The question is what happens on the way to the answer.

Pi separates responsibilities: parameter estimation, routing, session state, interpretation, and presentation, so they can be examined separately without collapsing into a single reply.

It makes parts of the system’s posture visible and provides a working environment for testing when an assistant should proceed, ask, hold, or offer a scaffold.