Introducing πππππππΈππππππΌπππππ’(): multi-turn conversations with context and intent understanding
β smart caching: vectorizes user intent from recent messages (π πππππ πππ£π)
β choose your own model for intent extraction
β cacheMode: "default" | "refresh", TTL, and threshold are fully customizable
β onStepFinish callback (cacheScore, step, userIntention, etc.)