Debugging multi-agent AI is hard. LangGraph gives you a span tree — but when something goes wrong you're still manually walking backwards to find which agent caused the cascade.
Rifft automates that walk. It finds the first handoff where state went wrong, classifies the failure mode (tool loop, handoff mismatch, context overflow), and shows you the causal chain.
Two lines to add to any LangGraph project.
Posture Corrector
