Save Costs in RAG Workflows with Q&A Tool & Multiple Models

This n8n workflow demonstrates a smart approach to implementing Retrieval-Augmented Generation (RAG) while minimizing costs. It starts with a Form Trigger node where users upload PDF or CSV files containing custom knowledge. These files are then processed through OpenAI Embeddings to create vector representations, stored in a Simple Vector Store for quick retrieval.

The core innovation is the Question and Answer (Q&A) Tool, which allows agents to query the vector store efficiently. Instead of running expensive full agent chains for every interaction, the Q&A tool retrieves relevant chunks and generates responses, drastically reducing token usage and API costs. This is ideal for production RAG setups where scalability and budget matter.

Benefits include no-setup execution—just upload and chat—plus easy customization like adding custom agent prompts, alternative data sources (e.g., Google Drive), advanced ranking, or production vector stores. Use cases span customer support bots with docs, internal knowledge bases, or any AI agent needing persistent, low-cost knowledge access. Read n8n's RAG docs for deeper integration tips.