WhatsApp RAG Chatbot n8n Workflow: Supabase, Gemini Flash

Turn WhatsApp into a searchable AI knowledge assistant with RAG—using n8n, Supabase, and Gemini

This n8n workflow lets you build a WhatsApp-based RAG chatbot that answers questions by retrieving relevant information stored in Supabase. It processes messages in real-time, creates semantic embeddings with OpenAI embeddings, and generates friendly responses using Gemini 2.5 Flash.

What this workflow does

• WhatsApp trigger: A new WhatsApp message starts the workflow via webhook.
• Message check: The workflow uses a switch step to determine whether the user sent a query or a document upload.
• Document handling: It fetches the file URL from WhatsApp, converts the binary file to text, generates OpenAI embeddings for the content, and stores them in Supabase for semantic retrieval.
• Query handling: For user questions, it generates query embeddings with OpenAI and retrieves the most relevant context from Supabase.
• Answer generation: The workflow sends the retrieved context to Gemini 2.5 Flash to produce user-friendly answers and sends them back directly to WhatsApp.
• Modular design: Document ingestion and query handling are structured so you can split them for larger setups.

Use cases

• Answer FAQ questions in WhatsApp using your existing internal documents.
• Provide customer support responses based on policies, product manuals, or help articles.
• Let teams query internal knowledge by uploading documents through WhatsApp.
• Deploy a simple RAG chatbot without coding by configuring credentials and running the template.

Technical details

• Core nodes: WhatsApp trigger, switch, HTTP Request (file URL fetching), and an n8n LangChain agent node (n8nn8n-nodes-langchainagent).
• RAG stack: OpenAI embeddings for both document and query embeddings + Supabase for storage and retrieval.
• LLM: Gemini 2.5 Flash to generate responses from retrieved context.
• Requirements: Configure Supabase and WhatsApp API credentials before running.