The origin story
Long before Netflix existed I had a simple dream: every movie, song, audiobook and book I loved, available instantly at home — on my terms, with no subscription and nobody else deciding what I could watch. This is what that turned into.
VHS
Tapes that took over floors and walls.
DVD
Entire walls of shelving, eventually.
ZIP & HDD
The burning era. I saw the future and jumped to files.
The vault
LeoneNAS 7 and LeoneMini-2. A media empire for family and close friends.
The lab
LeoneNAS 9. Same instinct, pointed at AI.
It started with VHS tapes that took over floors and walls. Then came DVDs, which eventually demanded entire walls of shelving. When the DVD burning era arrived, I saw the future and made the jump to digital files stored on ZIP disks and hard drives. That decision, more than twenty years ago, set me on the path of building something most people now pay monthly for.
Today that vision lives on LeoneNAS 7 — our 100 TB raw media vault — coupled with LeoneMini-2, the dedicated AV1 media engine. Thousands of movies, TV shows, albums, audiobooks, comics and books, perfectly organized and instantly streamable to any device, anywhere. No algorithms. No ads. No data harvested.
The same mindset that drove me from VHS to a self-hosted media library — own it, understand it, improve it — is exactly why I built LeoneNAS 9 for AI.
These machines have become both a playground and a classroom. Not just for me, but for my sons.
While most of the world rushed to use AI through cloud services, I chose to build it myself. Over the last year and a half I have been building, breaking and rebuilding AI servers because I wanted to understand these systems from the ground up rather than just consume them.
Artificial intelligence is the most significant technological shift since the internet. The people who understand how it works at the infrastructure level will have a real advantage. LeoneNAS 9 is my dedicated AI laboratory. By running powerful models locally instead of shipping everything to corporate clouds, I keep complete sovereignty over my data, my conversations and my knowledge.
The deepest purpose is my sons. Both already show strong technical aptitude, and I want them to grow up with an intimate, hands-on understanding of AI — how to run it, how to modify it, and how to trust it correctly. This project gives them a living classroom in systems engineering, open-source architecture and local AI orchestration.
It's about building something that belongs to us — a foundation of knowledge and independence they can inherit and build upon for the rest of their lives.
This dual-node engine is the heart of the family media world. LeoneNAS 7 holds the 100 TB raw vault and runs core backup and Caddy proxy services, while LeoneMini-2 provides dedicated, ultra-efficient AV1 hardware transcoding for high-concurrency streaming.
LeoneMini-2 handles high-efficiency hardware AV1 and 4K transcoding so everyone can stream on any device without buffering or high energy cost.
LeoneNAS 7 keeps 100 TB raw of vault media on Seagate Enterprise drives with mirrored NVMe boot.
Dedicated 5 GbE and bonded 2.5 GbE links keep playback smooth even with several family streams running at once.
This is my dedicated AI powerhouse — a UGREEN NASync iDX6011 Pro — and the home of Alister, my private AI assistant and digital twin. Every large language model, every personal knowledge base and every heavy generative task runs here on bare-metal hardware under Proxmox VE 9.2.5.
I spent over a year and a half testing, configuring and optimizing local AI infrastructure because I wanted to understand this technology from the inside out. I want my sons to grow up knowing how these systems actually work at the hardware and model-architecture levels.
132 TOPS — LeoneNAS-9 across the A2 and the Intel Arc NPU speed up local LLM inference and token generation so responses feel natural. 64 GB of LPDDR5X-8400 provides the memory bandwidth needed to load large models with high context windows. Proxmox VE 9.2.5 virtualizes Ollama, OpenWebUI, AnythingLLM and the local RAG pipelines cleanly, with hardware GPU passthrough.
Everything below starts closed. Open whichever part you actually care about.
LLM inference & interactive chat
The core thinking engine
Large language models are AI systems trained on enormous amounts of text. "Inference" is the process of the model actually thinking and generating a response in real time.
Running inference locally on LeoneNAS 9 means all of that computation happens in my house. Nothing is sent to OpenAI, Google or Anthropic. The conversations stay completely private.
Ollama + Llama models
Llama is a family of powerful open-source models from Meta. Ollama makes it easy to download and run them locally with strong reasoning and conversation quality.
Open WebUI
A private, feature-rich chat interface wired directly to Ollama. Voice, document processing, multi-user access and RAG — advanced local AI the whole family can use.
RAG — retrieval augmented generation
Giving the AI a personal memory
Normal LLMs have no knowledge of your specific life or documents. RAG fixes that by searching your private files first, then feeding the most relevant pieces to the model before it answers. The result is grounded, accurate and personal.
Meet Alister
My personal AI twin on LeoneNAS 9, powered by Ollama and AnythingLLM. It knows my projects, business documents, family stories and homelab knowledge, and answers questions about them accurately. Operator access only — not public, by choice.
Semantic search & knowledge discovery
Meaning, not just keywords
Traditional search looks for exact word matches. Semantic search understands meaning and intent. On LeoneNAS 9 the GPU accelerates building "embeddings", so searches across thousands of documents return almost instantly — even when the exact words don't match.
Generative AI
Creating new content privately
Beyond answering questions, generative models can create images, artwork and event graphics. On LeoneNAS 9 we run local models such as Stable Diffusion and Flux, so everything stays private and under our control — no third-party accounts, no data leaving the house.
This infrastructure represents years of learning and building, and a commitment to ownership and privacy. It isn't just servers. It's a foundation I'm building for my sons, and for anyone who cares about controlling their own digital life.