Google Gemini Chat Model and Redis integration

Overview This project is an AI-powered hotel receptionist built using n8n, designed to handle guest queries automatically through WhatsApp. It integrates Google Gemini, Redis, MySQL, and Google Sheets via LangChain to create an intelligent conversational system that understands and answers booking-related questions in real time. A standout feature of this workflow is its AI model-switching system — it dynamically assigns users to different Gemini models, balancing traffic, improving performance, and reducing API costs. How It Works WhatsApp Trigger The workflow starts when a hotel guest sends a message through WhatsApp. The system captures the message text, contact details, and session information for further processing. Redis-Based Model Management The workflow checks Redis for a saved record of the user’s previously assigned AI model. If no record exists, a Model Decider node assigns a new model (e.g., Gemini 1 or Gemini 2). Redis then stores this model assignment for an hour, ensuring consistent routing and controlled traffic distribution. Model Selector The Model Selector routes each user’s request to the correct Gemini instance, enabling parallel execution across multiple AI models for faster response times and cost optimization. AI Agent Logic The LangChain AI Agent serves as the system’s reasoning core. It: Interprets guest questions such as: “Who checked in today?” “Show me tomorrow’s bookings.” “What’s the price for a deluxe suite for two nights?” Generates safe, read-only SQL SELECT queries. Fetches the requested data from the MySQL database. Combines this with dynamic pricing or promotions from Google Sheets, if available. Response Delivery Once the AI Agent formulates an answer, it sends a natural-sounding message back to the guest via WhatsApp, completing the interaction loop. Setup & Requirements Prerequisites Before deploying this workflow, ensure the following: n8n Instance** (local or hosted) WhatsApp Cloud API** with messaging permissions Google Gemini API Key** (for both models) Redis Database** for user session and model routing MySQL Database** for hotel booking and guest data Google Sheets Account** (optional, for pricing or offer data) Step-by-Step Setup Configure Credentials Add all API credentials in n8n → Settings → Credentials (WhatsApp, Redis, MySQL, Google). Prepare Databases MySQL Tables Example: bookings(id, guest_name, room_type, check_in, check_out) rooms(id, type, rate, status) Ensure the MySQL user has read-only permissions. Set Up Redis Create Redis keys for each user: llm-user:<whatsapp_id> = { "modelIndex": 0 } TTL: 3600 seconds (1 hour). Connect Google Sheets (Optional) Add your sheet under Google Sheets OAuth2. Use it to manage room rates, discounts, or seasonal offers dynamically. WhatsApp Webhook Configuration In Meta’s Developer Console, set the webhook URL to your n8n instance. Select message updates to trigger the workflow. Testing the Workflow Send messages like “Who booked today?” or a voice message. Confirm responses include real data from MySQL and contextual replies. Key Features Text & voice support** for guest interactions Automatic AI model-switching** using Redis Session memory** for context-aware conversations Read-only SQL query generation** for database safety Google Sheets integration** for live pricing and availability Scalable design** supporting multiple LLM instances Example Guest Queries | Guest Query | AI Response Example | |--------------|--------------------| | “Who checked in today?” | “Two guests have checked in today: Mr. Ahmed (Room 203) and Ms. Priya (Room 410).” | | “How much is a deluxe room for two nights?” | “A deluxe room costs $120 per night. The total for two nights is $240.” | | “Do you have any discounts this week?” | “Yes! We’re offering a 10% weekend discount on all deluxe and suite rooms.” | | “Show me tomorrow’s check-outs.” | “Three check-outs are scheduled tomorrow: Mr. Khan (101), Ms. Lee (207), and Mr. Singh (309).” | Customization Options 🧩 Model Assignment Logic You can modify the Model Decider node to: Assign models based on user load, region, or priority level. Increase or decrease TTL in Redis for longer model persistence. 🧠 AI Agent Prompt Adjust the system prompt to control tone and response behavior — for example: Add multilingual support. Include upselling or booking confirmation messages. 🗂️ Database Expansion Extend MySQL to include: Staff schedules Maintenance records Restaurant reservations Then link new queries in the AI Agent node for richer responses. Tech Stack n8n** – Workflow automation & orchestration Google Gemini (PaLM)** – LLM for reasoning & generation Redis** – Model assignment & session management MySQL** – Booking & guest data storage Google Sheets** – Dynamic pricing reference WhatsApp Cloud API** – Messaging interface Outcome This workflow demonstrates how AI automation can transform hotel operations by combining WhatsApp communication, database intelligence, and multi-model AI reasoning. It’s a production-ready foundation for scalable, cost-optimized, AI-driven hospitality solutions that deliver fast, accurate, and personalized guest interactions.

This n8n template demonstrates one approach to achieve a more natural and less frustration conversations with AI agents by reducing interrupts by predicting the end of user utterances. When we text or chat casually, it's not uncommon to break our sentences over multiple messages or when it comes to voice, break our speech with the odd pause or umms and ahhs. If an agent replies to every message, it's likely to interrupt us before we finish our thoughts and it can get very annoying! Previously, I demonstrated a simple technique for buffering each incoming message by 5 seconds but that approach still suffers in some scenarios when more time is needed. This technique has no arbitrary time limit and instead uses AI to figure out when its the agent's turn based on the user's message, allowing for the user to take all the time they need. How it works Telegram messages are received but no reply is generated for them by default. Instead they are sent to the prediction subworkflow to determine if a reply should be generated. The prediction subworkflow begins by checking Redis for the current user's prediction session state. If this is a new "utterance", it kicks off the "predict end of utterance" loop - the purpose of which is to buffer messages in a smart way! New users message can continue to be accepted by the workflow until enough is collected to allow our prediction classifier to determine the end of the utterance has been reached. The loop is then broken and the buffered chat messages are combined and sent to the AI agent to generate a response and sent to the user via the telegram node. The prediction session state is then deleted to signal the workflow is ready to start again with a new message. How to use This system sits between your preferred chat platform and the AI agent so all you need to do is replace the telegram nodes as required. Where LLM-only prediction isn't working well enough, consider more traditional code-based checking of heuristics to improve the detection. Ideally you'll want a fast but accurate LLM so your user isn't waiting longer than they have to - at time of writing Gemini-2.5-flash-lite was the fastest in testing but keep a look out for smaller and more powerful LLMs in the future. Requirements Gemini for LLM Redis for session management Telegram for chat platform

AI Comprehensive Research on User's Query with Gemini and Web Search What is this? Perform comprehensive research on a user's query by dynamically generating search terms, querying the web using Google Search (by Gemini) , reflecting on the results to identify knowledge gaps, and iteratively refining its search until it can provide a well-supported answer with citations. (like Perplexity) This workflow is a reproduction of gemini-fullstack-langgraph-quickstart in N8N. The gemini‑fullstack‑langgraph‑quickstart is a demo by the Google‑Gemini team that showcases how to build a powerful full‑stack AI agent using Gemini and LangGraph How It Works Generate Query 💬 generates one or more search queries tasks based on the User's question. uses Gemini 2.0 Flash Web Research 🌐 execute web search tasks using the native Google Search API tool in combination with Gemini 2.0 Flash. Reflection 📚 Identifies knowledge gaps and generates potential follow-up queries. Setup Configure API Credentials: Create Google Gemini(PaLM) Api Credential using you own Gemini key Connect the credential with three nodes: Google Gemini Chat Model and GeminiSearch and reflection Configure Redis Source: prepare a Redis service that can be accessed by n8n Create Redis Crediential and connect it with all Redis node Customize Try using different Gemini models. Try modifying the parameters number_of_initial_queries and max_research_loops. Why use Redis? Use Redis as an external storage to maintain global variables (counter, search results, etc.) This workflow contains a loop process, which need global variables (as State in LangGraph). It is difficult to achieve global variables management without external storage in n8n.

🤖 Human-like Evolution API Agent with Redis & PostgreSQL This production-ready template builds a sophisticated AI Agent using Evolution API that mimics human interaction patterns. Unlike standard chatbots that reply instantly to every incoming message, this workflow uses a Smart Redis Buffering System. It waits for the user to finish typing their full thought (text, audio, or image albums) before processing, creating a natural, conversational flow. It features a Hybrid Memory Architecture: active conversations are cached in Redis for ultra-low latency, while the complete chat history is securely stored in PostgreSQL. To optimize token usage and maintain long-term coherence, a Context Refiner Agent summarizes the conversation history before the Main AI generates a response. ✨ Key Features Human-like Buffering:** The agent waits (configurable time) to group consecutive messages, voice notes, and media albums into a single context. This prevents fragmented replies and feels like talking to a real person. Hybrid Memory:* Combines Redis (Hot Cache) for speed and PostgreSQL* (Cold Storage) for permanent history. Context Refinement:** A specialized AI step summarizes past interactions, allowing the Main Agent to understand long conversations without exceeding token limits or increasing costs. Multi-Modal Support:** Natively handles text, audio transcription, and image analysis via Evolution API. Parallel Processing:** Manages "typing..." status and session checks in parallel to reduce response latency. 📋 Requirements To use this workflow, you must configure the Evolution API correctly: Evolution API Instance: You need a running instance of Evolution API. Configuration Guide N8n Community Node: Install the Evolution API node in your n8n instance. n8n-nodes-evolution-api Database: A PostgreSQL database for chat history and a Redis instance for the buffer/cache. AI Models: API keys for your LLM (OpenAI, Anthropic, or Google Gemini). ⚙️ Setup Instructions Install the Node: Go to Settings > Community Nodes in n8n and install n8n-nodes-evolution-api. Credentials: Configure credentials for Redis, PostgreSQL, and your AI provider (e.g., OpenAI/Gemini). Database Setup: Create a chat_history table in PostgreSQL (columns must match the Insert node). Redis Connection: Configure your Redis credentials in the workflow nodes. Global Variables: Set the following in the "Global Variables" node: wait_buffer: Seconds to wait for the user to stop typing (e.g., 5s). wait_conversation: Seconds to keep the cache alive (e.g., 300s). max_chat_history: Number of past messages to retrieve. Webhook: Point your Evolution API instance to this workflow's Webhook URL. 🚀 How it Works Ingestion: Receives data via Evolution API. Detects if it's text, audio, or an album. Smart Buffering: Holds the execution to collect all parts of the user's message (simulating a human reading/listening). Context Retrieval: Checks Redis for the active session. If empty, fetches from PostgreSQL. Refinement: The Refiner Agent summarizes the history to extract key details. Response: The Main Agent generates a reply based on the refined context and current buffer, then saves it to both Redis and Postgres. 💡 Need Assistance? If you’d like help customizing or extending this workflow, feel free to reach out: 📧 Email: [email protected] 🔗 LinkedIn: John Alejandro Silva Rodríguez

🤖 Multi-Agent AI Telegram Bot for Service Businesses Transform your salon/service business with this streamlined Telegram automation system featuring Claude integration, zero-setup database management, and intelligent conversation handling. How It Works Core Architecture Claude MCP Integration** - Direct connection to Claude Sonnet 4 via Model Context Protocol Streamlined 2-Agent System** - Booking Agent and Admin Agent (simplified from 5 for better reliability) GPT-5 Mini Primary** with Gemini 2.5 Flash backup for cost-effective processing Multi-Media Support** - Handles text, voice (Whisper transcription), images, and PDFs with cost extraction Smart Acknowledgments** - "One moment…", "Let me check availability…" during processing Rate Limiting & Spam Protection** - Configurable limits (default: 100 msg/hour) with professional UX Zero-Setup Database Management Autonomous Airtable Creation** - Bot creates all necessary tables automatically Complete CRUD Operations** - Create, edit, delete services and settings via Telegram Dynamic Business Configuration** - Modify hours, pricing, services conversationally Friend Booking Support** - "Book for my friend Sarah" functionality Setup Steps Telegram Business API setup (detailed instructions included) Airtable Base ID extraction (store in Redis or hardcode - recommended) Google Calendar integration for scheduling Redis for caching, rate limiting, and conversation memory MCP Server deployment for Claude integration Telegram for notifications Key Features Booking Management Booking Limit Control** - Default 6 appointments per customer (configurable in workflow) Service Name Matching** - GPT-5 Nano workflow for cost-optimized service recognition 24-Hour Advance Reminders** - Automatic Telegram reminders sent at 8 PM Conversation Handling Conversation Memory** maintains context across interactions Error Resilience** with backup models and graceful failure handling Use Cases Perfect for: Salons Spas Clinics Consulting services Any appointment-based business Complete business setup happens through conversational commands - no manual database configuration required. Transform your service business with intelligent automation powered by AI and Telegram integration.

🤖Multi-Agent AI WhatsApp Bot for Service Businesses Transform your salon/service business with this streamlined WhatsApp automation system featuring Claude integration, zero-setup database management, and intelligent conversation handling. How it works Claude MCP Integration** - Direct connection to Claude Sonnet 4 via Model Context Protocol Streamlined 2-Agent System** - Booking Agent and Admin Agent (simplified from 5 for better reliability) GPT-5 Mini Primary** with Gemini 2.5 Flash backup for cost-effective processing Multi-Media Support** - Handles text, voice (Whisper transcription), images, and PDFs with cost extraction Smart Acknowledgments** - "One moment...", "Let me check availability..." during processing Rate Limiting & Spam Protection** - Configurable limits (default: 100 msg/hour) with professional UX Zero-Setup Database Management Autonomous Airtable Creation** - Bot creates all necessary tables automatically Complete CRUD Operations** - Create, edit, delete services and settings via WhatsApp Dynamic Business Configuration** - Modify hours, pricing, services conversationally Friend Booking Support** - "Book for my friend Sarah" functionality Set up steps WhatsApp Business API** setup (detailed instructions included) Airtable Base ID** extraction (store in Redis or hardcode - recommended) Google Calendar** integration for scheduling Redis** for caching, rate limiting, and conversation memory MCP Server** deployment for Claude integration Whatsapp** for notifications Key Features Booking Limit Control** - Default 6 appointments per customer (configurable in workflow) Service Name Matching** - GPT-5 Nano workflow for cost-optimized service recognition 24-Hour Advance Reminders** - Automatic WhatsApp reminders sent at 8 PM Conversation Memory** maintains context across interactions Error Resilience** with backup models and graceful failure handling Perfect for salons, spas, clinics, consulting services, or any appointment-based business. Complete business setup happens through conversational commands - no manual database configuration required.