Multi-Modal Multilingual WhatsApp Customer Support Agent - n8n Workflow

Companies offering customer support via WhatsApp who need multilingual capabilities (English and Roman Urdu).
Users looking for advanced examples of Retrieval Augmented Generation (RAG) within an n8n workflow.
Automation specialists deploying complex, multi-modal n8n templates integrating various AI services (OpenAI, OpenRouter, n8n LangChain nodes).
Businesses requiring a hands-off customer service solution that utilizes existing knowledge bases in Google Docs.

This sophisticated n8n workflow provides a complete, 24/7 customer support solution on the WhatsApp platform. It solves the critical challenge of handling diverse incoming message formats (text, voice notes, and images) and responding accurately across multiple languages (specifically English and Roman Urdu, based on the prompt configuration).

The core of this n8n automation leverages a powerful AI Agent configured with Anthropic's Claude 3.5 Sonnet (via OpenRouter) and connected to a Google Docs knowledge base. This RAG setup ensures that all responses are accurate and sourced from official company documentation. Every n8n node is meticulously placed to create a resilient, human-like conversational experience, maintaining chat history using the Simple Memory n8n node, making this one of the most advanced n8n templates available for customer support.

The process begins immediately upon message receipt via the dedicated n8n trigger:

1. Trigger and Classification: The WhatsApp Trigger n8n trigger initiates the n8n workflow upon receiving any message. A Check Input Type Switch n8n node immediately classifies the incoming message as Text, Voice, or Image.


2. Multi-Modal Processing:

1. Intelligent AI Agent: All processed text converges on the AI Agent n8n node. This agent operates using the OpenRouter Chat Model (Claude Sonnet 4) and utilizes the Simple Memory n8n node to maintain conversation history, crucial for contextual replies. The extensive system prompt dictates tone (Spartan, Human-like), language matching, and constraints.


2. Knowledge Retrieval: If the query requires information, the AI Agent uses its assigned Get a document in Google Docs tool to search the configured Google Doc (acting as the knowledge base) and retrieve the precise details needed.


3. Response Delivery: The final, verified answer generated by the AI is passed to the Respond with Text n8n node, which securely sends the multilingual response back to the original customer's WhatsApp number, completing the n8n workflow cycle.

To implement this n8n workflow, follow these steps:

1. Import: Copy the provided n8n workflow JSON and paste it into your n8n instance.


2. WhatsApp Credentials: You will need two WhatsApp Business API connections:

1. AI Credentials:

1. Google Docs Tool: Create an Google Docs OAuth2 credential and configure the Get a document in Google Docs n8n node with the ID of your company's knowledge base document.


2. HTTP Header Auth: Configure the Header Auth account credentials used by the Download Audio/Image HTTP Request n8n node to ensure media files are retrieved correctly from the WhatsApp API. This setup ensures this complex n8n template runs smoothly.

This n8n workflow utilizes several key specialized n8n node types:

WhatsApp Trigger: The starting n8n trigger. It listens for messages updates from the configured WhatsApp Business account, providing the initial data to the n8n workflow.
Check Input Type (Switch): A core logic n8n node that routes the workflow based on the message type (text, audio, or image). This is essential for multi-modal handling.
WhatsApp (Media URL Get): Used in both the Image and Audio branches to call the WhatsApp API and retrieve the temporary URL needed to download the media file.
HTTP Request (Download Media): Fetches the actual binary data of the audio or image file, utilizing header authentication to access the secure URL provided by WhatsApp.
OpenAI (Analyze Image/Transcribe Audio): An n8n node used for high-level media processing. It uses the analyze operation for images (GPT-4o) and the transcribe operation for audio (Whisper), preparing content for the AI Agent.
Set (Prompt Formatters): Simple n8n node used to structure the input text (text field) that will be fed into the LangChain AI Agent, combining data from various steps.
Simple Memory (LangChain): An n8n node that implements conversation memory, keyed by the customer's phone number, allowing the AI Agent to remember previous interactions within the ongoing n8n workflow.
OpenRouter Chat Model (LangChain): Provides the underlying LLM (Claude Sonnet 4) that powers the conversational intelligence of the n8n node AI Agent.
Google Docs Tool: Configured as a RAG tool within the n8n workflow, enabling the AI Agent to retrieve and utilize specific, accurate company information from the Google Doc.
AI Agent (LangChain): The central processing n8n node. It manages the detailed conversational logic, uses the provided memory and tools, and adheres to the multilingual and constraints defined in its system prompt.

• WhatsApp (Respond with Text): The final action n8n node, sending the AI-generated response back to the recipient.