Build a WhatsApp AI Auto-Reply with n8n, WOZTELL, and Google Gemini

Route WhatsApp inbound messages into n8n via webhook, filter and fetch conversation context, generate replies with Gemini, and send back through WOZTELL

Who

Teams with an n8n instance, a WOZTELL WhatsApp channel, and a Gemini API key

You get

When a customer sends a WhatsApp text message, the system automatically generates a context-aware professional reply and sends it

You need

n8n

Steps

1. Import the template

   In the n8n workflow editor, import template ID 13641 'Build an AI agent for WhatsApp with WOZTELL and Google Gemini'. After import, you will see nodes including Webhook, Filter, Get conversation history, Edit Fields, AI Agent, Google Gemini Chat Model, and Send responses.

2. Configure the WOZTELL webhook

   Activate the workflow, then copy the Webhook node's Production URL and paste it into your WOZTELL inbound webhook settings. Under Event Subscriptions, select Inbound Messages. In the Member Metadata field, enter: Member Metadata: liveChat; Member Path: member.botMeta.liveChat.

   Setup guidance · Source text。WOZTELL inbound webhook configuration UI

   Copy prompt

   After activating this workflow, copy the **Production URL** and paste it into your WOZTELL inbound webhook settings.
   
   Check how to configure the webhook [here](https://support.woztell.com/portal/en/kb/articles/web#Create_Webhooks).
   
   Select **Inbound Messages** in the Event Subscriptions.
    
   In the Member Metadata field, fill in the following details:
   Member Metadata: **liveChat**
   Member Path: **member.botMeta.liveChat**

3. Connect WOZTELL credentials and set history permissions

   Add a woztellCredentialApi credential in n8n. Generate a WOZTELL access token with these permissions: channel:list, channel:getBasicInfo, member:getConversations, bot:sendResponses, member:getDetails. If you already have a Channel API token, you may reuse it. The Get conversation history node defaults to maxResults of 10.

   Setup guidance · Source text。WOZTELL credential setup and access token generation guide

   Copy prompt

   Reads up to 100 messages for the account to provide context to the AI.
   
   Set up your WOZTELL credentials. To generate the access token, follow the step-by-step guide [here]( https://support.woztell.com/portal/en/kb/articles/access-token#Access_Token_Generation).
   
   Required permissions:
   - channel:list
   - channel:getBasicInfo
   - member:getConversations
   - bot:sendResponses
   - member:getDetails
   ###
   For the WOZTELL Channel API field, follow [this guide](https://support.woztell.com/portal/en/kb/articles/access-token-channels-in-woztell#How_to_Create_an_Access_Token_for_a_Channel_in_WOZTELL) to create a new token. If you have already generated a token, you may reuse it.

4. Connect Google Gemini credentials

   Add a googlePalmApi credential in n8n and enter your Gemini API key. The AI Agent node calls the Google Gemini Chat Model sub-node with a default maxOutputTokens of 1000. You may adjust token limits or switch to another supported model.

   Setup guidance · Source text。Google Gemini credential configuration node note

   Copy prompt

   Google Gemini Chat Model provides the AI language model used to generate replies.
   
   Connect your Google Gemini API key. Follow the official n8n documentation [here](https://docs.n8n.io/integrations/builtin/cluster-nodes/sub-nodes/n8n-nodes-langchain.lmchatgooglegemini/).
   
   You can adjust token limits or switch to another supported model if needed.

5. Activate and test

   Activate the workflow. Send a text message to the WhatsApp number linked to WOZTELL. If eventType=inbound, type=text, and liveChat=false, the system retrieves conversation history, formats it, passes it to the AI Agent, generates a plain-text-only reply via Gemini, and sends it back through WOZTELL. Messages that fail the filter are silently ignored.

You are a Professional Conversation Analyst and Reply Generator Agent. Analyze the provided conversation snippets. Generate a single, concise, and contextually appropriate reply that directly addresses the last [Customer] statement in a professional tone, (Output **ONLY the plain text** of the reply).