Facebook Messenger AI customer service chatbot with Google Gemini + conversation memory - free template

Built with n8n
, ,
1

Sharing a workflow I built for handling Facebook Messenger customer service with an AI backend. Posting here in case it helps others who are trying to set up Messenger bots on n8n.

Template: Build a Facebook Messenger customer service AI chatbot with Google Gemini | n8n workflow template

How it works

The flow has three stages:

  1. Webhook & validation - Receives both GET (verification handshake) and POST (messages) from Facebook. Filters out non-text events and blocks echo messages (the bot’s own replies coming back through the webhook - this trips up a lot of people).

  2. AI processing - Extracts user_id, page_id, message text and token via Set Context. Sends “seen” + “typing” indicators immediately so users get feedback. Runs the message through Gemini Flash with a 10-message memory window per user.

  3. Format & send - Strips markdown from the response (Messenger doesn’t render it), truncates at 1900 chars, sends via Graph API.

A few things I learned the hard way

  • Echo filtering is essential. Facebook sends your bot’s own messages back to your webhook. Without filtering sender ID vs page ID, you get infinite loops.
  • Simple Memory doesn’t persist in queue mode. Works fine for testing. For stable production you need Redis or Postgres-backed sessions.
  • Gemini Flash is fast enough. With a tight system prompt, response times stay under 2 seconds - feels natural in Messenger.
  • Token management is a separate problem. I built a dedicated workflow for long-lived page token refresh: Get long-lived Facebook Page access tokens and subscribe Messenger webhook fields via Graph API | n8n workflow template

What this template doesn’t cover (upgrade paths)

Happy to answer questions about setup or any issues you’ve run into with the Facebook Graph API on n8n.

2

did you test echo filtering with concurrent webhooks? we had some issues on high-volume bots where the filter timing wasnt quite right.

1 Like
4

This workflow is really just a learning playground, since it’s built with Simple Memory. But if your n8n is already on v1.113 or higher, you’re in luck, you can grab a similar workflow called “Auto‑reply to Instagram DMs with an AI chatbot and Google Gemini history” from my official library. And the best part? It’s totally free right here: nguyenthieutoan | n8n Creator .

That one is battle‑tested to handle hundreds of users at the same time, because each person gets their own unique ID and the conversation history is neatly stored in a data table.

And if you want to level up, I also offer premium versions for just a bit over $10, packed with extra goodies like:

  • Handling interrupted messages smoothly

  • Automatically pausing when an admin jumps in

So you can start with the free template for fun and experiments, and upgrade later if you want more professional features.