Rifft

Rifft provides developers with tools to monitor and diagnose issues in AI agent systems built on frameworks such as CrewAI and LangGraph. By automatically capturing spans for tool calls and state mutations, the platform reconstructs the sequence of events leading to breakdowns and highlights the initial point where incorrect data was introduced. The system organizes detected problems into a taxonomy of failure classes including unverified outputs, repeated tool invocations, schema mismatches during agent transitions, and excessive context accumulation. Users can then fork a run at any identified span, apply adjustments, and compare results side by side to validate improvements before deployment. Integration requires minimal code changes through a simple wrapper that works across supported stacks and open telemetry setups. This approach reduces the time spent reviewing extensive logs by focusing attention on causal chains rather than surface-level error messages.