does anyone know how good the knowledge base in Vapi.ai works? Is it a RAG knowledge base? --> If I have a big knowledge base, can I use the vapi knowledge base, or is it a better idea to implement a function that then calls a RAG knowledgebase in Vectorshift for example?