An SDK that preserves voice-agent session state across mid-call interrupts and cross-agent handoffs

Real-time voice agents built on OpenAI Realtime, Pipecat, LiveKit, Vapi, Retell, or Bland lose all in-flight state the moment a user barges in while a tool call is executing or when a call is routed to a second agent. The audio pipeline cancels or replays; the tool result is orphaned or replayed out of order; the new agent starts cold. Developers currently stitch together their own checkpoint-and-replay wrappers, which are fragile, untested at scale, and re-built from scratch for every framework. This SDK provides a framework-agnostic middleware layer that checkpoints tool-call state before and during execution, reconciles barge-in events with in-flight tool results, serializes full conversational context for cross-agent handoffs, and recovers dropped or stale audio sessions from the last clean checkpoint. It ships as a drop-in adapter for every major voice-agent framework and exposes a recovery-event observability stream so teams can measure and tune recovery quality in production.