AI Infrastructure Stack

A production-ready infrastructure stack combining DevOps tooling with AI-powered automation — semantic search, intelligent workflows, local LLM processing, and enterprise monitoring.

Architecture

graph TD
    A[Traefik Reverse Proxy] --> B[n8n Workflows]
    A --> C[Knowledge Base]
    A --> D[Monitoring Stack]

    B --> E[AI Agents]
    E --> F[Knowledge Manager]
    E --> G[Task Intelligence]
    E --> H[Content Intelligence]
    E --> I[Monitoring Analytics]

    J[LEANN Vector DB] --> K[OpenAI Embeddings]
    J --> L[Redis Cache]

    P[Ollama LLM] --> Q[Qwen 2.5 3B]
    B --> P
    E --> J

    M[Prometheus] --> N[Grafana]
    M --> O[AlertManager]

Loading

Components

Service	Purpose	Port
Traefik	Reverse proxy + auto-SSL	80/443
n8n	Workflow automation + agent orchestration	5678
Ollama	Local LLM inference (Qwen 2.5 3B, 100% offline)	11434
LEANN	Vector database for semantic search	3001
Redis	LEANN query cache (99.97% latency reduction)	6379
Prometheus	Metrics collection	9090
Grafana	Dashboards (13+ pre-configured)	3000
AlertManager	Alert routing	9093
Portainer	Container management UI	9000

AI Agents

Four autonomous agents run on n8n, each with LEANN semantic search integration:

Agent	Endpoint	Function
Knowledge Manager	POST /webhook/km-agent-test	PARA-categorized knowledge ingestion with contextual analysis
Task Intelligence	POST /webhook/task-intelligence	Task analysis, execution planning, and conditional auto-execution
Content Intelligence	POST /webhook/content-intelligence	Web scraping with duplicate detection via semantic similarity
Monitoring Analytics	POST /webhook/monitoring-analytics	Proactive infrastructure health monitoring via Prometheus

LEANN Semantic Search

Vector database with OpenAI embeddings and Redis caching:

# Search
curl -X POST http://localhost:3001/search \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"query": "AI agent implementation", "index": "vault", "top_k": 5}'

Performance: 15ms cached vs 44s uncached (99.97% improvement at >90% cache hit rate).

Local LLM (Ollama)

• Model: Qwen 2.5 3B Instruct (Q4_0, 2.2GB RAM)
• Response time: 2–3 seconds
• Privacy: Fully local — no external API calls

Quick Start

git clone https://github.com/fredericosanntana/ai-infrastructure-stack.git
cd ai-infrastructure-stack
cp docker-compose/.env.example docker-compose/.env
# Edit .env with your configuration
docker compose -f docker-compose/docker-compose.yml up -d

Prerequisites

• Docker & Docker Compose
• OpenAI API key (for LEANN embeddings)
• Domain with DNS configured (for Traefik SSL)

Monitoring

• Prometheus collects metrics from all services and containers
• Grafana provides 13+ dashboards: infrastructure overview, AI agent performance, LEANN cache metrics, Redis stats
• AlertManager routes alerts via configurable channels

Performance

Metric	Value
Full stack memory	~4GB
Average CPU	<10%
Uptime	99.9%+
Agent response	1–6s depending on agent
Storage baseline	~2GB + data volumes

Documentation

• Architecture Overview
• AI Agents System
• LEANN Vector Database
• Ollama LLM Integration
• AI Agents Deployment Guide

License

MIT