AI-powered email replies with spam filtering & FAQ lookup using GPT-4o mini & Pinecone

This n8n workflow automates the handling of incoming emails. It detects and filters out spam, searches a knowledge base (FAQ) stored in a Pinecone vector database, and sends a reply using Gmail — all powered by an AI model (GPT-4o mini).

How It Works

1. Receiving Emails

• The Gmail Trigger node checks a Gmail inbox every hour.

• When a new email arrives, it starts the workflow.

1. Fetching Full Email Content

• The get_message node retrieves all the details of the message: sender, subject, text, message ID, etc.

1. Spam Filtering

• The Spam checker node uses GPT-4o mini to classify the email as either "spam" or "no spam".

• It detects not only classic spam but also automated messages (e.g. from Google or Microsoft).

• If marked as "spam", the workflow ends and nothing is processed.

1. Conditional Filter

• The If node checks the spam result.

• Only "no spam" emails proceed to the AI Agent.

1. AI-Based Reply

• The AI Agent node generates a response based on:

  • The email content

  • A system prompt defining the assistant’s behavior (polite, professional, under the name “Total AI Solutions”)

  • Information retrieved from the Pinecone Vector Store, which contains FAQs

• The AI is instructed to always check the vector store before replying.

• The AI prepares both the subject and the body of the reply.

1. Sending the Reply

• The Gmail node sends the reply to the original sender.

• It uses the original email's ID to keep the thread intact.

1. Language Model

• The OpenAI Chat Model node provides GPT-4o mini as the language engine for generating responses.

1. Memory Support

• The Simple Memory node maintains short-term context, helpful in multi-turn conversations.

1. Knowledge Base (FAQ)

• The Pinecone Vector Store node connects to a Pinecone index (faqmattabott) containing vectorized FAQ content.

• Vectors are created using the Embeddings OpenAI node.