AI Workflows with N8N, PostgreSQL, and Ollama

A comprehensive AI-powered system that combines N8N's visual workflow builder with PostgreSQL's vector database capabilities and local Ollama models to create intelligent agents and document processing workflows.

🏗️ Architecture Overview

This project demonstrates the integration of three powerful technologies:

• N8N: Visual AI workflow builder for creating sophisticated AI agents and processing pipelines
• PostgreSQL: Vector database for efficient document storage and semantic search
• Ollama: Local LLM runtime for privacy-focused AI processing

🚀 Available Workflows

The flows/ directory contains pre-built N8N workflows for various AI tasks:

1. Document Q&A Workflow

• File: Document Q&A.json
• Purpose: Question-answering system for uploaded documents
• Components: Chat input, prompt templates, Ollama model integration
• Use Case: Interactive document querying with natural language

2. Docling Processing Workflow

• File: Docling Processing.json
• Purpose: Advanced document parsing and processing using Docling
• Components: Document ingestion, text splitting, export functionality
• Use Case: Structured data extraction from various document formats

3. News and Web Search Workflow

• File: News and Web Search.json
• Purpose: AI-powered web content aggregation and analysis
• Components: News search, web scraping, agent-based processing
• Use Case: Real-time information gathering and summarization

🛠️ Technology Stack

N8N Integration

• Visual Workflow Builder: Drag-and-drop interface for creating AI pipelines
• Multi-Agent Support: Orchestrate complex AI interactions
• Component Library: Pre-built components for common AI tasks
• API Deployment: Export workflows as RESTful endpoints
• Code Access: Full Python customization capabilities

PostgreSQL Vector Database

• Vector Extension: pgvector for high-dimensional similarity search
• Document Storage: Efficient storage of document embeddings
• Semantic Search: Fast cosine similarity queries
• Scalable Architecture: IVFFlat indexes for production performance

Ollama Integration

• Local Models: Run LLMs locally without API costs
• Privacy First: No data sent to external services
• Model Flexibility: Support for Llama, Mistral, and custom models
• Embedding Generation: Snowflake Arctic Embed for semantic search

📁 Project Structure

ai_workflows/
├── flows/                          # N8N workflow definitions
│   ├── Docling Processing.json     # Document processing pipeline
│   ├── Document Q&A.json          # Question-answering workflow
│   └── News and Web Search.json   # Web content aggregation
├── postgres/                       # Database schema and setup
└── ollama/                       # Ollama Docker Files

🗄️ Database Schema

The PostgreSQL database includes optimized tables for document storage and retrieval:

-- Documents table with vector embeddings
CREATE TABLE documents (
    id SERIAL PRIMARY KEY,
    filename TEXT NOT NULL,
    content TEXT NOT NULL,
    metadata JSONB NOT NULL,
    embedding vector(1536),
    created_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP
);

-- Document chunks for improved retrieval
CREATE TABLE document_chunks (
    id SERIAL PRIMARY KEY,
    document_id INTEGER REFERENCES documents(id) ON DELETE CASCADE,
    chunk_index INTEGER NOT NULL,
    content TEXT NOT NULL,
    embedding vector(1536) NOT NULL,
    created_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP
);

🎯 Key Features

Document Processing

• Multi-format Support: PDF, text, and web content processing
• Intelligent Chunking: Optimal text segmentation for embedding
• Metadata Extraction: Automatic document metadata capture
• Vector Embeddings: High-quality semantic representations

AI Agent Capabilities

• Natural Language Understanding: Advanced query comprehension
• Contextual Responses: Relevant answer generation
• Multi-turn Conversations: Maintained conversation state
• Tool Integration: External API and database access

User Interface

• Intuitive Upload: Drag-and-drop document ingestion
• Real-time Search: Instant semantic query results
• Responsive Design: Mobile-friendly interface
• Progress Tracking: Visual feedback for processing status

🔧 Development Setup

Prerequisites

• PostgreSQL 14+ with pgvector extension
• Python 3.10+ (for N8N)
• Ollama installed locally
• Docker (optional, for containerized deployment)

Installation

1. Clone the repository

   git clone <repository-url>
   cd ai_workflows

2. Set up PostgreSQL

   # Create database
   createdb vectordb
   
   # Enable vector extension
   psql vectordb -c "CREATE EXTENSION IF NOT EXISTS vector;"

3. Install Ollama

   # Download from https://ollama.ai
   # Pull required models
   ollama pull llama3.2
   ollama pull snowflake-arctic-embed2

4. Install N8N

N8N Self Hosting

5. Configure environment variables

   DATABASE_URL=postgresql://postgres:postgres@localhost/vectordb
   OLLAMA_HOST=http://localhost:11434

Running the Application

1. Start PostgreSQL

   # Using Docker
   docker-compose up -d postgres

2. Start Ollama

   ollama serve
   # Runs at http://localhost:11434

3. Launch N8N

   docker compose up -d
   # Access at http://localhost:5678

API Integration

• Workflows expose RESTful endpoints automatically
• Ollama provides local model inference
• Real-time processing with streaming responses
• Error handling and retry mechanisms

🚀 Deployment

Docker Deployment

# Build and run all services
docker-compose up -d

Production Considerations

• Database Scaling: Connection pooling and read replicas
• Caching Strategy: Redis for session and result caching
• Load Balancing: Nginx for traffic distribution
• Monitoring: Application performance and error tracking

🔮 Future Enhancements

• Multi-modal Processing: Image and audio document support
• Advanced Analytics: Usage metrics and insights
• Workflow Templates: Pre-built industry-specific solutions
• Collaboration Features: Multi-user workflow sharing
• Enhanced Security: Authentication and authorization

📚 Resources

• N8N Documentation
• PostgreSQL pgvector Extension
• Ollama Documentation
• Snowflake Arctic Embed

---

Built with ❤️ using N8N, PostgreSQL, and Ollama