Dynamic Website Assistant with DeepSeek AI, Pinecone Vectorstore & Site-Based Routing

This workflow contains community nodes that are only compatible with the self-hosted version of n8n.

🚀 Overview
This workflow enables a powerful AI-driven virtual assistant that dynamically responds to website queries using webhook input, Pinecone vector search, and OpenAI agents — all smartly routed based on the source website.

🔧 How It Works

Webhook Trigger
The workflow starts with a Webhook node that receives query parameters: query: The user's question
userId: Unique user identifier
site: Website identifier (e.g., test_site)
page: Page identifier (e.g., homepage, pricing)

Smart Routing
A Switch node directs the request to the correct AI agent based on the site value. Each AI agent uses: OpenAI GPT-4/3.5 model
Pinecone vector store for context-aware answers
SQL-based memory for consistent multi-turn conversation

Contextual AI Agent
Each agent is customized per website using: Site-specific Pinecone namespaces
Predefined system prompts to stay in scope
Webhook context including page, site, and userId

Final Response
The response is sent back to the originating website using the Respond to Webhook node.

🧠 Use Case
Ideal for multi-site platforms that want to serve tailored AI chat experiences per domain or page — whether it’s support, content discovery, or interactive agents.

✅ Highlights
🧠 Vector search using Pinecone for contextual responses
🔀 Website-aware logic with Switch node routing
🔐 No hardcoded API keys
🧩 Modular agents for scalable multi-site support