Might be helpful for anyone using llms and want cost savings! https://quesma.com/blog/tau2-benchmark-improving-results-smaller-models/

In this video I show a new, faster way to build RAG agents in n8n using the Pinecone Assistant to handle the heavy lifting, no preprocessing pipeline, no manual chunking, no custom embedding flow. You simply drop your documents in and Pinecone takes care of ingestion, splitting, and indexing on the back end. Then we wire it up in n8n to get accurate answers with trustworthy, page-level citations and exact text quotes pulled straight from your knowledge base. If you’re a beginner, this is the easiest path I’ve found to stand up a production-ready RAG agent; if you’re more advanced, it’s a huge time-saver for rapid prototyping. Follow along step-by-step as I connect n8n to Pinecone Assistant, load docs, test grounded responses, and return clean sources you can show to clients or stakeholders, all with a simple, no-code build that you can replicate in minutes.