👉 n8n No Limits Access :

https://www.hostg.xyz/SHHAd

Hébergement VPS n8n | Automatisation sécurisée et évolutive

coupon -15% : DRFIRAS15

👉 Formation n8n en Français

Formation n8n en Français : Automatisation & Agents IA

coupon -20% : LANCEMENT : HERE

What you will build

A multimodal AI assistant that lives on your WhatsApp Business number. It understands text, voice notes, images and videos, remembers the conversation per contact, can look things up on Wikipedia, and replies straight back in the chat.

<aside> 🎯

This page is a filming script. Everything is copy-paste ready so you can rebuild the whole workflow live, node by node, without opening the finished template.

</aside>

Flow diagram

WhatsApp Trigger
   └─> Split out messages
         └─> Configuration (Set)
               └─> Route by message type (Switch)
                     ├─ 0 Audio ─> Get audio URL ─> Download audio ─> Transcribe audio with OpenAI ─┐
                     ├─ 1 Video ─> Get video URL ─> Download video ─> Describe video with Gemini ─> Extract video description ─┤
                     ├─ 2 Image ─> Get image URL ─> Download image ─> Analyze image with GPT ─┤
                     └─ 3 Text  ─> Get text message ─────────────────────────────────────────┤
                                                                                              ▼
                                                                              Normalize user message
                                                                                              ▼
                                                                                        AI Agent
                                                          (OpenAI Chat Model + Conversation Memory + Wikipedia)
                                                                                              ▼
                                                                              Format reply for WhatsApp
                                                                                              ▼
                                                                                 Send WhatsApp reply

Why this design

Decision Reason
GPT-5 for chat, vision and transcription One provider, one API key, best reasoning quality. Whisper handles voice notes natively.
Gemini only for video Gemini is one of the very few models that accepts raw video input. The free tier is enough for testing.
Text messages skip the LLM pre-processing The original template summarised text before the agent — that wastes tokens and loses nuance.
A single Configuration node A beginner changes the model, the memory size or the persona in ONE place.
Format reply node WhatsApp does not render Markdown. Without cleaning, users see **bold** and ## headers as raw text.

Prerequisites & credentials

Credential Type Used by
WhatsApp OAuth account whatsAppTriggerApi WhatsApp Trigger
WhatsApp account whatsAppApi Get audio/video/image URL, Download audio/video/image, Send WhatsApp reply
OpenAi account openAiApi Transcribe audio, OpenAI Vision Model, OpenAI Chat Model
Google Gemini(PaLM) Api account googlePalmApi Describe video with Gemini

1. WhatsApp Business Cloud (Meta)

  1. Go to developers.facebook.com → My Apps → Create App → type Business.