If you cannot infer the property from the user's message, do not call any tools and instead ask for more information in the chat.
If the person requests to be contacted, do not call any tools and instead return a response indicating that someone will reach out to them.
Use a friendly, helpful tone.Bonus: add a confidence check before routing. One step further: you can add a lightweight confidence check before routing happens at all. Use a Basic LLM Chain node to evaluate the incoming message and return a structured confidence score — essentially asking the model to rate how clearly the message maps to a known domain. Then feed that score into an If or Switch node: if confidence is high enough, route to the appropriate Pinecone Assistant Tool node as normal; if it's too low, short-circuit to a clarifying response before any retrieval happens. It's a cheap gate that prevents a whole class of bad retrievals before they start.Adapting the patternThe vacation rental scenario is a stand-in for any workflow where you have meaningfully different knowledge domains:Franchise locations (each location gets its own Pinecone Assistant)Client accounts in an agency context (client A's docs stay in client A's assistant)Support tiers (enterprise documentation doesn't pollute SMB tier answers)Internal teams (runbooks, HR policies, and sales playbooks shouldn't share a knowledge base)Multi-tenant applications (each end-user or customer gets their own assistant so one tenant's data is never retrievable by another)The signal that you need domain separation: when a wrong answer can be explained by "it pulled from the wrong part of the knowledge base," and that part is logically unrelated to the question.The full n8n workflowThe n8n workflow template for the vacation rental implementation is available here. It includes the full file upload path (Google Drive Trigger → Google Drive node → Pinecone Assistant node) and the chat path (Chat Trigger → AI Agent node → Pinecone Assistant Tool nodes → response).To adapt it, the main things to change are the routing conditions in the AI Agent's system message and the number of domain-specific Pinecone Assistant Tool nodes. The rest of the structure stays the same.Wrap upRAG isn't a monolithic system—it's a pattern you can shape to match how your data is actually organized. If your knowledge base covers multiple distinct domains, separating them into dedicated Pinecone Assistants and routing queries to the right one before retrieval is a straightforward way to improve accuracy, make debugging easier, and keep scaling manageable. The n8n workflow template gives you a working starting point; the routing logic is the main thing to make your own. The payoff: guests get the right answer for their property, every time—and you can debug, update, or scale any one domain without touching the rest.You can find the rest of our n8n resources here:
