WhatsApp assistant that reads voice notes, photos and video

Newversion 1 · updated 2026-11-13

WhatsApp assistant that reads voice notes, photos and video, picture 1

An n8n workflow that answers messages sent to a WhatsApp business number. A Gemini model turns a voice note, a photo or a video into words, and an AI agent with a short memory and a Wikipedia tool writes the reply.

WhatsApp assistant that reads voice notes, photos and video is an n8n workflow in the Productivity category, listed by Mindket. It is free. This is version 1, last updated on 2026-11-13. Download it on this page or copy its JSON, then import it into an n8n canvas. Nothing runs until you activate it in n8n.

At a glance

Runs in

n8n Cloud: import the file there, and n8n gives the address that Meta calls
Your own n8n server: the same file and steps, once the server has a public HTTPS address
Another AI: swap the Gemini nodes for any model that takes audio, images and video

Platform

n8nWorkflow automation, in n8n’s cloud or on your own server

Connects 3 apps

How it works

Trigger

A WhatsApp message

It starts each time someone sends a message to your WhatsApp business number, and there is no form:

A text, a voice note, a photo or a video, with a caption if there is one

Actions

The message is checked for its type: voice note, video, photo or text
A voice note or a video is downloaded and sent to Gemini, which writes down or describes it
A photo is downloaded and a Gemini chain describes it and copies any text in it
A text is shortened, and a message with no text gets a short note
An AI agent with a short memory and a Wikipedia tool writes the answer, and WhatsApp sends it back

Result

The person gets a text answer on WhatsApp, whatever kind of message they sent

Integrations

What you need before you start.

Chat app

Delivers each message, serves the media files and sends the answer.

NeedsA Meta business app, a Client ID and Client Secret, an access token and a Business Account ID

Google Gemini

AI model

Writes down voice notes, describes photos and videos, and writes the answers.

NeedsA Google Gemini API key

Wikipedia

Encyclopedia

Gives the agent facts when a question needs one.

NeedsNothing, it has no key

Workflow check

Checked by Mindket on 13 Nov 2026

Steps: 19 steps, starting on a Whats App event
AI: 3 steps call an AI model: AI Agent, Image Explainer and Text Summarizer
Sign-ins: Signs in to: WhatsApp Business and Google Gemini
API keys: No API key is left in the file
Step types: Mindket’s checker reads all 9 of its step types

Read from the workflow’s file, not from a run.

About

What it is

This n8n workflow turns your WhatsApp business number into an AI assistant. When someone writes, the workflow checks what kind of message it is. A voice note is written out, a photo is described and a video is summarized, all by a Google Gemini model. An AI agent then answers in the same chat, remembers the last few messages and can look facts up on Wikipedia. Twenty-four nodes do the work, beside eleven sticky notes.

Who it suits

Small teams that get questions on WhatsApp and want a first answer at any hour, such as shops, clinics, tutors and support desks. It is also a good start if you plan to build your own WhatsApp bot. Anyone with the number can write to it, so start with a test number.

Who uses it, and how

Industry	Who uses it	How they use it	Example
Retail	Small shops	Answer questions about products at night	A customer sends a photo of a broken part, and the bot describes it and asks for the model number.
Health	Clinics	Take a first message in the patient’s own words	A patient leaves a voice note, and the agent replies with the opening hours and asks for a day.
Education	Tutors	Give students a quick answer between lessons	A student photographs a page of homework and gets a first explanation.
Support	Help desks	Read what the customer sent before a person steps in	A user sends a short video of an error, and the team reads the description.

You need some n8n and a Meta app: import a file, make a WhatsApp business app and add four credentials.

Where to use it

Any n8n with a public HTTPS address can run it, because WhatsApp delivers each message to that address.

n8n Cloud: import the file there, and n8n gives the address that Meta calls
Your own n8n server: the same file and steps, once the server has a public HTTPS address
Another AI: swap the Gemini nodes for any model that takes audio, images and video

Setup

Download saves a zip with workflow.json and Mindket’s setup notes in your downloads folder. You need a Meta business app with WhatsApp, and a Google Gemini API key.

What you need first

Need	Detail
n8n	On n8n Cloud or on a server with a public HTTPS address
Meta	A developer account and a business app with the WhatsApp product, with its Client ID and Client Secret
WhatsApp	An access token and the Business Account ID of the number you answer from
Google Gemini	An API key from Google AI Studio

In n8n, create a workflow, open its menu, choose Import from File and pick workflow.json.
In WhatsApp Trigger, add a WhatsApp OAuth credential with your Meta app’s Client ID and Client Secret.
In Get Audio URL, Get Video URL, Get Image URL and Respond to User, add a WhatsApp credential with your access token and Business Account ID.
In Download Audio, Download Video and Download Image, choose that same WhatsApp credential.
In Respond to User, choose your number in the Phone Number ID list, in place of <YOUR_PHONE_NUMBER_ID>.
Add a Google Gemini (PaLM) API credential with your key. Choose it in the three Gemini chat model nodes and in Google Gemini Audio and Google Gemini Video.
Turn the workflow on, because WhatsApp only sends messages to an active workflow.
From your own phone, send the number a text, a voice note, a photo and a short video, and check each answer.

How to use it

Once the workflow is on, each message to the number starts one run, and one text answer comes back. The agent keeps the last five messages of each sender, until n8n restarts.

What to change for a different result

To change	Where
What the assistant is and how it speaks	The system message in the Options of `AI Agent`
The Gemini model	The three chat model nodes and the URL of `Google Gemini Audio` and `Google Gemini Video`
What Gemini does with a voice note or a video	The text in `Google Gemini Audio` and `Google Gemini Video`
How many messages it remembers	The window size in `Window Buffer Memory`
What else it can do	Add tools to `AI Agent`

Example

A clinic answers a voice note at night.

A patient records a short voice note at 11 pm and asks when the clinic opens on Saturday. The workflow downloads the audio and Gemini writes down what was said. The agent replies in a few lines with the hours from its instructions, and asks which day suits the patient. The reply is only as good as the instructions, so the clinic puts its real hours there.

For newcomers and experts

If you are new. n8n draws a workflow as boxes joined by lines, and each box, a node, does one job. WhatsApp sends a message to the workflow, and the workflow sends one back. Gemini is Google’s AI model, and it can listen to audio and look at pictures. Try it with your own phone before you share the number.

For experts. The switch has four outputs, with text as the fallback, and every path ends in one Set node that gives the agent the same three fields.

Where to look closer

Node	What it does	What to know
`WhatsApp Trigger`	Registers the webhook and receives each call	A call holds a list of messages, and a delivery receipt has none
`Redirect Message Types`	Routes by `type` to audio, video, image or text	Anything else, such as a sticker, goes down the text path
`Google Gemini Audio`	Sends the audio inline and keeps the transcript	The request holds the file as base64, and the total is capped at 20 MB
`Image Explainer`	Describes the photo and copies any text	It uses the Gemini chat model and the image binary
`AI Agent`	Writes the answer with memory and Wikipedia	Version 1.6 of the agent node

Good to know

The original copy named a Gemini model that Google no longer lists, and anyone who has the number can use the bot.

Model: the original copy named gemini-1.5-pro-002 in five places. Google’s model list, read on 12 November 2026, has no Gemini 1.5 model, and its page lists gemini-3.8-flash as Stable, with text, image, video, audio and PDF as input. This copy names gemini-3.8-flash everywhere
Open bot: the workflow has no list of allowed senders, so anyone who has the number can spend your Gemini quota
Window: Meta’s guide, read on 12 November 2026, allows a free-form reply only inside 24 hours of the user’s last message. The bot answers at once, so it stays inside
Size: Meta lists 16 MB for audio and video and 5 MB for an image. Google caps one request at 20 MB in total, so a long video can fail, and the workflow does not check
Short link: WhatsApp’s media address lasts 5 minutes, and the download follows at once
Files on disk: the Gemini calls read $input.item.binary.data.data, which holds the file only when n8n keeps files in memory. If it stores them on disk or in S3, add an Extract From File step that moves the file to a base64 string
No text: a sticker or a location gets a short note instead of an empty prompt, so the agent can still answer

Usernames: Meta says a user with a username can have no phone number in the call, so from is missing and the reply fails. Mindket did not check whether n8n’s WhatsApp node can send to the ID that comes instead
Memory: it lives in the n8n process, so a restart clears it
Text only: every reply is text, whatever the person sent
Your data: each message, voice note, photo and video goes to Google, and the answer goes through Meta
Not run: Mindket did not call WhatsApp, Google or Wikipedia to test this file, and ran it in a test runner with stand-in answers
Credentials: the JSON holds none, and the phone number id of its author is gone

README and files

README

WhatsApp assistant that reads voice notes, photos and video

An n8n workflow that answers messages sent to your WhatsApp business number. A text goes straight to an AI agent. A voice note, a
photo or a video is first turned into words by a Google Gemini model, and the agent then answers in the same chat. The agent
keeps a short memory of each person's chat and can look facts up on Wikipedia.

What is inside

workflow.json: the workflow as n8n exports it, 24 working nodes and 11 sticky notes. It holds no credential.
README.md: these notes.

What we changed from the original copy

The original copy named the model gemini-1.5-pro-002 in five places: the three chat model nodes and the two HTTP calls to Gemini.
Google's model list, read on 12 November 2026, has no Gemini 1.5 model at all. All five now name gemini-3.8-flash, which Google's
page lists as Stable, with text, image, video, audio and PDF as input.
The two HTTP calls to Gemini took the file type from the file extension. They now read the mime_type that WhatsApp returns in
Get Audio URL and Get Video URL, drop anything after a semicolon, and use the extension only when it is missing.
Text Summarizer stopped with an empty prompt when a message had no text, such as a sticker or a location. It now gets a short
note saying what kind of message arrived, so the agent can answer.
The phone number id of the author's number is now <YOUR_PHONE_NUMBER_ID>, and the agent prompt has its typos fixed and says how
to use Wikipedia.
The memory key whatsapp-tutorial-<number> is now whatsapp-<number>.
The sticky notes are rewritten and have no outside link. Every credential reference and the webhook ids are removed.

Before you start

An n8n instance with a public HTTPS address. WhatsApp delivers each message to it.
A Meta developer account and a business app with the WhatsApp product added. The credentials page of n8n, read on 12 November 2026,
says the app needs a privacy policy address and a terms of service address. It must also be switched to Live mode and reviewed for a full launch.
Two WhatsApp credentials in n8n. The trigger takes the app's Client ID (the Meta app ID) and Client Secret (the app secret). The send
and media steps take an Access Token and a Business Account ID.
A Google Gemini API key from Google AI Studio.

Set it up

In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
In WhatsApp Trigger, add a WhatsApp OAuth credential with your Meta app's Client ID and Client Secret.
In Get Audio URL, Get Video URL, Get Image URL and Respond to User, add a WhatsApp credential with your access token and your
Business Account ID.
In Download Audio, Download Video and Download Image, choose that same WhatsApp credential.
In Respond to User, replace <YOUR_PHONE_NUMBER_ID> by choosing your number in the Phone Number ID list.
Add a Google Gemini (PaLM) API credential with your key. Choose it in Google Gemini Chat Model, Google Gemini Chat Model1,
Google Gemini Chat Model2, Google Gemini Audio and Google Gemini Video.
Turn the workflow on. WhatsApp only sends messages to an active workflow.
From your own phone, send the business number a text, then a voice note, a photo and a short video. Check each answer.

How a run goes

WhatsApp Trigger receives a call from WhatsApp. Split Out Message Parts makes one item for each message. A call that holds no
message, such as a delivery receipt, gives no item and ends there.
Redirect Message Types sends the item down one of four paths by its type: audio, video, image, or anything else as text.
For audio and video, Get Audio URL or Get Video URL asks WhatsApp for the file's address, Download Audio or Download Video
fetches the file, and Google Gemini Audio or Google Gemini Video sends it to Gemini inline. Format Response1 or Format Response
keeps the text that comes back.
For an image, Get Image URL and Download Image fetch the file, and Image Explainer asks Gemini to describe it and copy any text.
For text, Get Text passes the message on and Text Summarizer shortens it. A message with no text gets a plain note instead.
Get User's Message collects the type, the text or description, the caption and the sender's number.
AI Agent answers with Google Gemini Chat Model, Window Buffer Memory and the Wikipedia tool.
Respond to User sends the answer to the number that wrote.

Change the look

The agent's instructions are in the Options of AI Agent. Change them to fit your business, and add tools to the agent to book
appointments, look up orders or check documents.
The model is in the three chat model nodes and in the URL of the two Gemini HTTP calls.
How many earlier messages the agent remembers is the window size in Window Buffer Memory, five by default.
What Gemini does with a voice note or a video is the text in Google Gemini Audio and Google Gemini Video.

Good to know

Anyone who has the number can write to it. Each message costs Gemini tokens, and the workflow has no list of
allowed senders.
Meta's guide, read on 12 November 2026, says a user message opens a 24-hour customer service window. Free-form replies are allowed inside
it and only template messages outside it. This workflow answers at once, so it stays inside.
WhatsApp's media address works for 5 minutes and needs the access token, so the download follows at once. Meta lists 16 MB as the
limit for audio and video and 5 MB for an image.
Google's page on audio, read on 12 November 2026, caps one request at 20 MB in total, files included, so a long video can fail. The
workflow does not check a file's size.
The Gemini calls read the file with $input.item.binary.data.data. If your n8n keeps files on disk or in S3, that field holds a
reference and not the file. Add an Extract From File step that moves the file to a base64 string.
Meta's page on business-scoped user IDs, read on 12 November 2026, says a user who turned on a username can have no phone number in
the call. The call then carries from_user_id and no from. This workflow reads from for the memory key and for the reply, so
such a person gets no answer. Mindket did not check whether n8n's WhatsApp node can send to that ID.
The memory lives in the n8n process, so a restart clears it. One session is kept for each sender's number.
A voice note, a photo or a video is described by Gemini, and the agent only reads the description. The reply is always text.
The agent is on node version 1.6 and the Gemini chat models on version 1. Newer versions exist, and this copy was not moved to them.
Your data: every message, and every voice note, photo or video, goes to Google, and the answer goes through Meta.
Not run: Mindket did not call WhatsApp, Google or Wikipedia to test this file. It ran in a test runner with stand-in answers, and a text,
a voice note, a video, a photo and a sticker each reached a reply.
The JSON holds no credential.

Files

NOTICE.md  625 B
README.md  7 KB
workflow.json  26 KB

3 files, 33 KB, downloaded as a zip.

What’s new

Version 1 · 2026-11-13

Reviews

No ratings yet

Write a review

No reviews yet.

Similar in Productivity

See all in Productivity

Frontend Slides

zarazhangrui · Skill

Builds HTML presentations from scratch or from a PowerPoint file and shows visual style previews to pick from, for people who want web slides with no CSS or Ja…

New32.3k

IPTV

iptv-org · Kit

Collects publicly available live TV channels from many countries into playlists that any video player with live streaming can open. Made for viewers and for de…

New144k

upscaler

uwg · Kit

Holds a set of image upscaling models gathered from the OpenModelDB community, for people who want to enlarge low-resolution pictures with AI on their own comp…

New

WhatsApp assistant that reads voice notes, photos and video

Overview

At a glance

Platform

How it works

Trigger

Actions

Result

Integrations

WhatsApp

Google Gemini

Wikipedia

Workflow check

About

What it is

Who it suits

Where to use it

Setup

How to use it

Example

For newcomers and experts

Good to know

README and files

WhatsApp assistant that reads voice notes, photos and video

What is inside

What we changed from the original copy

Before you start

Set it up

How a run goes

Change the look

Good to know

What’s new

Reviews

Write a review

More from Mindket

Huashu Design

Guizang PPT Skill

Claude Code Skills & Plugins

Similar in Productivity

Frontend Slides

IPTV

upscaler