Automated Gmail support agent with Gemini 2.5, RAG & Cohere reranking

AI Email Support Agent with RAG & Cohere Reranking

Transform your inbox into an intelligent support system: automatically detect new emails, retrieve relevant knowledge from Pinecone, rerank with Cohere for precision, generate contextual replies using Gemini AI, and respond—all while maintaining conversation history.

What It Does

This workflow triggers on incoming Gmail messages, leverages a LangChain agent with PostgreSQL memory for context, queries a Pinecone vector store (RAG) enhanced by Cohere reranking and OpenAI embeddings, crafts personalized responses via Gemini 2.5, and auto-replies to keep support flowing.

Key Features

• Gmail Integration - Real-time polling for new emails every minute
• RAG with Pinecone - Retrieves top 10 relevant docs from "agency-info" index as agent tool
• Cohere Reranking - Boosts retrieval accuracy by reordering results semantically
• Persistent Memory - Postgres chat history keyed by email ID for ongoing threads
• Gemini-Powered Agent - Handles queries with custom system prompt for agency support
• Seamless Auto-Reply - Sends formatted text responses directly in Gmail

Perfect For

• Agencies: Automate client FAQs on services, pricing, and ownership
• Support Teams: Scale responses without losing conversation context
• Small Businesses: Handle inquiries 24/7 with AI-driven accuracy
• Developers: Prototype RAG agents with vector stores and rerankers
• Marketers: Personalize outreach replies based on knowledge base
• Consultants: Quick, informed answers from internal docs

Technical Highlights

Built on n8n's LangChain ecosystem, this workflow highlights:

• Trigger-to-response pipeline with polling and webhooks
• Hybrid retrieval: Embeddings + vector search + semantic reranking
• Stateful agents with database-backed memory for multi-turn chats
• Multi-provider setup: OpenAI (embeddings), Cohere (rerank), Google (LLM)
• Scalable for production with configurable topK and session keys

Setup Instructions

Prerequisites

• n8n instance with LangChain nodes enabled
• Accounts for: Gmail (OAuth2), OpenAI (API key), Cohere (API key), Google Gemini (API key), Pinecone (API key and index), Postgres (database connection, e.g., Neon or Supabase)

Required Credentials

Gmail OAuth2

1. Enable Gmail API in Google Cloud Console
2. Create OAuth2 credential in n8n with scopes: https://www.googleapis.com/auth/gmail.readonly, https://www.googleapis.com/auth/gmail.send

OpenAI API

1. Get API key from platform.openai.com
2. Add as OpenAI credential in n8n

Cohere API

1. Sign up at cohere.com
2. Copy API key to n8n Cohere credential

Google Gemini API

1. Generate key at https://aistudio.google.com/
2. Add as Google PaLM credential in n8n (compatible with Gemini)

Pinecone API

1. Create index "agency-info" with dimension 1024
2. Add API key to n8n Pinecone credential

Postgres

1. Set up database (e.g., Neon/Supabase) with a table for chat history
2. Add connection details (host, database, user, password) to n8n Postgres credential

Configuration Steps

1. Import the workflow JSON into your n8n instance
2. Assign all required credentials to the respective nodes
3. Populate the Pinecone "agency-info" index with your knowledge base documents (use a separate upsert workflow or Pinecone dashboard)
4. Customize the tableName in the Postgres Memory node if needed (default: "email_support_agent_")
5. Adjust the agent's system prompt or topK retrieval if required for your use case
6. Activate the workflow and test by sending a sample email to trigger it

Troubleshooting

• No trigger firing: Verify Gmail scopes and polling interval
• Empty retrieval: Check Pinecone index population, dimensions (must be 1024), and document embeddings
• Rerank errors: Ensure Cohere API key is valid and has sufficient quota
• Memory issues: Confirm Postgres connection and that sessionKey uses email ID

Perfect for deploying hands-off email automation. Import, connect credentials, and activate!