Build a PDF-based RAG system with OpenAI, Pinecone and Cohere reranking

This workflow contains community nodes that are only compatible with the self-hosted version of n8n.

This workflow provides a complete, ready-to-use template for a Retrieval-Augmented Generation (RAG) system. It allows you to build a powerful AI chatbot that can answer questions based on the content of PDF documents you provide, using a modern and powerful stack for optimal performance.

Good to know

• Costs: This workflow uses paid services (OpenAI, Pinecone, Cohere). Costs will be incurred based on your usage. Please review the pricing pages for each service to understand the potential expenses.
• Video Tutorial (Bahasa Indonesia): For a step-by-step guide on how this workflow functions, you can watch the accompanying video tutorial here: N8N Tutorial: Membangun Chatbot RAG dengan Pinecone, OpenAI, & Cohere

How it works

This workflow operates in two distinct stages:

1. Data Ingestion & Indexing:

• It begins when a .pdf file is uploaded via the n8n Form Trigger.
• The Default Data Loader node processes the PDF, and the Recursive Character Text Splitter breaks down the content into smaller, manageable chunks.
• The Embeddings OpenAI node converts these text chunks into vector embeddings (numerical representations).
• Finally, the Pinecone Vector Store node takes these embeddings and stores (upserts) them into your specified Pinecone index, creating a searchable knowledge base.

2. Conversational AI Agent:

• A user sends a message through the Chat Trigger.
• The AI Agent receives the message and uses its VectorDB tool to search the Pinecone index for relevant information.
• The Reranker Cohere node refines these search results, ensuring only the most relevant context is selected.
• The user's original question and the refined context are sent to the OpenAI Chat Model (gpt-4.1), which generates a helpful, context-aware answer.
• The Simple Memory node maintains conversation history, allowing for natural, multi-turn dialogues.

How to use

Using this workflow is a two-step process:

1. Populate the Knowledge Base: First, you need to add documents. Trigger the workflow by using the Form Trigger and uploading a PDF file. Wait for the execution to complete. You can do this for multiple documents.
2. Start Chatting: Once your data has been ingested, open the Chat Trigger's interface and start asking questions related to the content of your uploaded documents.

The Form Trigger is just an example. Feel free to replace it with other triggers, such as a node that watches a Google Drive or Dropbox folder for new files.

Requirements

To run this workflow, you will need active accounts and API keys for the following services.

• OpenAI Account & API Key:
  • Function: Powers text embedding and the final chat generation. Required for the Embeddings OpenAI and OpenAI Chat Model nodes.
• Pinecone Account & API Key:
  • Function: Used to store and retrieve your vector knowledge base. Required for the Pinecone Vector Store and VectorDB nodes. You also need to provide your Pinecone Environment.
• Cohere Account & API Key:
  • Function: Improves the accuracy of your chatbot by re-ranking search results for relevance. Required for the Reranker Cohere node.

Customising this workflow

This template is a great starting point. Here are a few ways you can customize it:

• Change the AI Personality: Edit the System Message in the AI Agent node to change the bot's behavior, tone, or instructions.
• Use Different Models: You can easily swap the OpenAI model for another one (e.g., gpt-3.5-turbo for lower costs) in the OpenAI Chat Model node.
• Adjust Retrieval: In the VectorDB tool node, you can modify the Top K parameter to retrieve more or fewer document chunks to use as context.
• Automate Ingestion: Replace the manual Form Trigger with an automated one, like a node that triggers whenever a new file is added to a specific cloud storage folder.