Building your first WhatsApp chatbot

Product brochure PDF → vector knowledge base → WhatsApp trigger → AI sales agent → auto-reply

Who

n8n users who want to build a WhatsApp chatbot

You get

An AI sales assistant that answers product questions by retrieving from a vector store knowledge base

You need

n8n · WhatsApp Business

Steps

1. Import product brochure and build vector knowledge base

   Use the HTTP Request node to fetch a product brochure PDF (the template example uses the Yamaha 2024 powered loudspeakers brochure), then extract text with the Extract from File node. Load the text with Default Data Loader, split into chunks of 2000 characters with Recursive Character Text Splitter, generate embeddings via OpenAI Embeddings (text-embedding-3-small), and store in the In-Memory Vector Store (memoryKey: whatsapp-75, mode: insert, clearStore: true). This step only needs to run once; re-run when updating the brochure.

2. Configure WhatsApp trigger and message filtering

   Add a WhatsApp Trigger node (listen for messages updates), connect to a Switch node named Handle Message Types to filter by message type: only text messages flow to the Supported branch, others go to Not Supported. Non-text messages receive a fixed reply: I'm unable to process non-text messages. Please send only text messages. Thanks! After configuration, activate the workflow; for self-hosted n8n ensure WhatsApp can reach your server.

3. Set up the Sales AI Agent

   Add an AI Agent node after the Supported branch, configured as Agent with Tools. Connect an OpenAI Chat Model (gpt-4o-2024-08-06) as the language model; attach a Window Buffer Memory sub-node (sessionIdType: customKey, sessionKey: =whatsapp-75-{{ $json.messages[0].from }}) to isolate sessions per user; attach a Vector Store Tool sub-node (name: query_product_brochure, description: Call this tool to query the product brochure. Valid for the year 2024.) linked to the Product Catalogue vector store.

   Execution prompt · Source text。AI Sales Agent node System Message field (promptType: define)

   Copy prompt

   You are an assistant working for a company who sells Yamaha Powered Loudspeakers and helping the user navigate the product catalog for the year 2024. Your goal is not to facilitate a sale but if the user enquires, direct them to the appropriate website, url or contact information.
   
   Do your best to answer any questions factually. If you don't know the answer or unable to obtain the information from the datastore, then tell the user so.

4. Reply to users and activate the workflow

   Connect the AI Agent output to a WhatsApp node's textBody field (expression: ={{ $json.output }}), with recipientPhoneNumber dynamically set from the trigger message sender. After configuration, click Test Workflow once to populate the vector store, then activate the workflow. Send a text message to the bound WhatsApp number to receive replies from the AI sales agent. Replace the PDF source, adjust the systemMessage, or switch the vector store backend as needed.

You are assisting with reproducing n8n template #2465 "Building your first WhatsApp chatbot". Please proceed in this order:
1. Confirm you have an n8n instance, WhatsApp Business credentials (whatsAppTriggerApi + whatsAppApi), and an OpenAI API key (openAiApi).
2. Import workflow.json or copy the template from the n8n template library.
3. Replace the URL in the get Product Brochure node with an actual product brochure PDF URL (the original template contains a placeholder YOUR_AWS_SECRET_KEY_HERE).
4. Replace the phoneNumberId in the WhatsApp nodes with your real WhatsApp Business phone number ID.
5. First click Test Workflow to run the vector store build part (once only).
6. Activate the workflow and test by sending a text message to the bound WhatsApp number.
7. If you need persistent storage, replace the In-Memory Vector Store with Qdrant/Pinecone and update the corresponding credentials.