Pi Durable gives long-running agents a way to recover after their process stops. Released alongside Pi 1.0, the experimental framework saves a checkpoint for each step. A replacement process opens the same storage and resumes unfinished work. Interrupted model requests are sent again; interrupted tools rerun only when declared safe to repeat. Otherwise, the model is told the call was interrupted and decides what to do next. The implementation account makes recovery behavior explicit, so resuming a conversation need not mean blindly repeating every action.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
