AI Lab Server Setup

Universal provisioning scripts for AI/DevOps lab environments on Ubuntu 24.04.

One script turns a fresh server into a fully configured AI lab with Docker, Ollama, Qdrant, Python venv, firewall, and SSH hardening. Optionally deploy a full platform stack — Dify, Flowise, n8n, Langfuse, and Traefik — with a single docker compose up. Works with any cloud provider or bare metal — not tied to a specific platform.

System Requirements

Resource	Minimum	Recommended
CPU	2 vCPU	4 vCPU
RAM	4 GB (base only)	8 GB (full platform stack)
Disk	40 GB	80 GB
OS	Ubuntu 24.04 LTS	Ubuntu 24.04 LTS

Hetzner Cloud equivalents:

Type	vCPU	RAM	Use Case
CPX22	2	4 GB	Base setup only (no Dify, no LLM models)
CPX32	4	8 GB	Full stack + small LLM (recommended)
CPX42	8	16 GB	Full stack + larger LLM models

Quick Start

Option A: Cloud-Init (Fully Automatic)

Pass examples/cloud-config.yml as user-data when creating a VM. Cloud-init clones this repo and runs setup.sh automatically.

# Copy and customize the template
cp examples/cloud-config.yml cloud-config.yml
nano cloud-config.yml  # add your SSH public key

# Create VM (example: Hetzner Cloud)
hcloud server create \
  --name ai-lab \
  --type cpx32 \
  --image ubuntu-24.04 \
  --user-data-from-file cloud-config.yml

Wait 3-5 minutes, then ssh lab@<server-ip>.

Option B: Clone and Run

ssh root@<server-ip>
git clone https://github.com/DenAV/ai-lab-server-setup.git /home/lab/ai-lab-server-setup
bash /home/lab/ai-lab-server-setup/setup.sh

Option C: Pipe Over SSH

ssh root@<server-ip> 'bash -s' < setup.sh

The script auto-clones the repo for config files if not running from a local copy.

What Gets Installed

Component	Version	Purpose
Docker Engine	latest	Container runtime
Ollama	latest	Local LLM inference
Qdrant	v1.12.1	Vector database
Python 3.12	system	Python environment with venv
UFW	system	Firewall (SSH, HTTP, HTTPS)
Fail2ban	system	Brute-force protection

Default Ollama models pulled in background: llama3.2, nomic-embed-text.

Files

File	Description
setup.sh	Main setup script — run on any fresh Ubuntu 24.04
config/fail2ban.conf	Fail2ban jail configuration
config/bash_aliases	Shell shortcuts for lab user
docker-compose.yml	AI platform stack (Dify, Flowise, n8n, Ollama, Qdrant, Langfuse, Traefik)
.env.example	Environment variables for docker-compose
scripts/generate-env.sh	Generate .env with auto-generated secrets (only domain + email needed)
scripts/validate.sh	Post-setup health check
scripts/collect-diagnostics.sh	Collect logs and configs into a zip for support
examples/cloud-config.yml	Cloud-init template (works with any provider)
docs/	Detailed setup guides and architecture decision records
docs/update-server.md	How to apply repo changes to a running server
TROUBLESHOOTING.md	Common issues and solutions

AI Platform Stack (Optional)

After running setup.sh, deploy the full platform stack:

cd ~/ai-lab-server-setup

# Generate .env with auto-generated secrets (only domain + email needed)
bash scripts/generate-env.sh example.com user@example.com

# Or interactively:
bash scripts/generate-env.sh

# Start all services
docker compose up -d

# View generated credentials
cat .secrets

Services included:

Service	Subdomain	Purpose	Guide
Traefik	—	Reverse proxy with automatic TLS	setup
Dify	dify.<domain>	AI application platform	setup
Flowise	flow.<domain>	Visual AI agent builder	setup
n8n	n8n.<domain>	Workflow automation	setup
Ollama	internal	Local LLM runtime	setup
Qdrant	internal	Vector database	setup
Langfuse	trace.<domain>	LLM observability	setup
Demo DB	internal	Shared PostgreSQL for demo projects	—

Deploying demo projects

The lab serves as infrastructure for AI demo presentations and quick agent assembly. Any project that uses Dify, n8n, Ollama, or PostgreSQL can be deployed on top of the running platform stack.

Common requirements for demo projects:

• External API keys (OpenAI, Telegram, etc.) — configure in platform UIs, not in .env
• HTTPS webhook URLs — provided by Traefik automatically
• PostgreSQL for chat/data logging — use the shared demo-db container (create a database per project)
• Dify apps and Knowledge Bases — created via Dify web UI

Each demo project should include its own deployment guide in a deploy/ directory with platform-specific instructions. See docs/demos/ for the general deployment workflow.

Configuration

Override defaults via environment variables before running setup.sh:

export LAB_USER="myuser"
export TIMEZONE="America/New_York"
export QDRANT_VERSION="v1.13.0"
export OLLAMA_MODELS="llama3.2 mistral nomic-embed-text"
./setup.sh

Variable	Default	Description
LAB_USER	lab	Non-root user to create
TIMEZONE	Europe/Berlin	Server timezone
QDRANT_VERSION	v1.12.1	Qdrant Docker image tag
OLLAMA_MODELS	llama3.2 nomic-embed-text	Models to pull (space-separated)

Validation

Check that everything is working:

~/ai-lab-server-setup/scripts/validate.sh
# or via alias:
lab-validate

Security

• Root SSH login disabled after setup
• Password authentication disabled (key-only)
• UFW firewall: only SSH (22), HTTP (80), HTTPS (443)
• Fail2ban protects SSH (3 attempts, 1h ban)
• All secrets in .env (gitignored, never committed)

WARNING: setup.sh disables root SSH access. Make sure you can login as the lab user before closing your root session.

License

MIT