"Welcome to Sovereign Seed"

My Personnal Setup

"Hardware Setup"

This setup is already strong enough to run models up to 7–8B parameters, and even test Mixture of Experts (MoE) architectures.

Looking Ahead

I’m planning to upgrade soon with a Geforce RTX 5090, but NVIDIA has introduced a Verified Priority Access program for the GeForce RTX 5090, limiting purchases to one card per household and available by invitation only. This shows how critical and in‑demand high‑end GPUs have become.

I’m also considering moving to Ryzen 64-bit for even more power.

---

"Base System Setup"

You can install both Pytorch and TensorFlow without any conflict — many developers use them side by side. I currently use both in my setup. Just make sure they are correctly linked to your NVIDIA drivers, CUDA, and cuDNN so your GPU can handle AI workloads efficiently.

⚙️ Essential Base Installations

Development & Compatibility

Python 3 + pip
Almost all open‑source AI frameworks rely on it.

Git
Essential for cloning repositories from Hugging Face or GitHub.

Build Essentials
(gcc, g++, make, etc. — required to compile certain modules)

---

"Software Stack"

Docker Compose: The Foundation of Your Local AI Setup

For any local AI project, the very first step is to install Docker Compose.

It’s the backbone of the infrastructure, acting like a server that runs multiple services together — databases, vector stores, APIs, and AI models — all seamlessly connected and working in harmony.

With Docker, you can create a complete AI setup from a single configuration file. An image is more than just software it’s a ready‑to‑run package that brings together everything needed, connecting services directly inside containers.

You can build your own images using a Dockerfile , but in most cases, Docker Hub already provides thousands of ready‑to‑use images for almost every need.

While Docker Compose may seem challenging at first, it quickly becomes the key to a smooth, efficient and modular AI setup.
👉 Start here: Install Docker Compose

For a deeper dive, I highly recommend this 20‑episode Docker training, which helped me truly understand and feel at ease with Docker.

N8n for workflow automation

A No‑Code Platform for AI Automation

n8n is an open‑source automation tool designed for people who want to build complex workflows without needing to write code. With its visual interface, you simply connect nodes together, each representing an action or a service to create powerful automation chains in minutes.

Visual Workflows, Zero Coding Instead of typing hundreds of lines of code, you just drag and drop nodes.

Want to connect an AI model with a vector database? → Just link the nodes.

Need speech‑to‑text, text‑to‑speech, or image generation? → There are nodes for that too.

Even advanced features like RAG (Retrieval‑Augmented Generation) and PostgreSQL integration are just a few clicks away.

It’s so intuitive that you’ll quickly find yourself creating workflows that look as good as they are useful.

Connect Everything With n8n, you can connect almost anything: Local AI models (Ollama, Hugging Face, etc.) Vector databases like Qdrant ChatGPT APIs and other cloud AI services STT/TTS systems for speech interaction PostgreSQL and other databases for structured data Even image processing tools If it has an API, you can plug it into n8n.

Why You’ll Love It No‑Code → Perfect for beginners who don’t know how to program. Extremely Visual → Easy to understand, easy to design. Endlessly Flexible → Works with AI, databases, APIs, and more. Addictive → Once you start building workflows, you won’t want to stop.

👉 Try n8n today — you’ll be hooked in no time.

PostgreSQL for structured data

PostgreSQL is a powerful, open‑source relational database system.

It stores and organizes structured data — like user information, logs, or application records — in a secure and reliable way.

Unlike vector databases such as Qdrant, which are designed for embeddings and semantic search, PostgreSQL focuses on structured data

This makes it ideal for:

Saving and querying user profiles Storing application settings and logs Managing large volumes of structured information that your AI may need

👉 For more details : PostgreSQL official website .

In short, Qdrant remembers meaning, while PostgreSQL remembers structure — and together, they form a complete memory system for your AI

Qdrant: The Vector Database for AI Memory

With Qdrant, you can create a private knowledge base that your AI truly understands. Here’s how it works, step by step:

Chunking your documents

You send text or documents into a chunking system.

The content is split into smaller, meaningful pieces (chunks), such as sentences or short paragraphs.

Creating embeddings

Each chunk is converted into an embedding (a vector) using a model like Ollama’s TextEmbed.

These embeddings capture the semantic meaning of the text, not just the words.

👉 For example: “car” and “vehicle” may look different, but in vector space they appear close, because they mean the same thing.

Storing in Qdrant

The embeddings are stored in Qdrant, creating your own semantic library of knowledge.

You can also add metadata (topics, tags, sources) for more targeted search later.

Asking questions

When you ask your AI a question, the question itself is transformed into an embedding.

Qdrant compares this query vector with the stored vectors and finds the closest matches.

Retrieving context

The most relevant chunks are returned to your AI.

Combined with structured context from PostgreSQL (user data, metadata, or logs), this gives your model a complete picture.

In short: Qdrant gives your AI semantic memory, while PostgreSQL adds structured context. Together, they make your AI smarter and more personalized.

👉 Download Qdrant Locally

Ollama: Powering Local AI with Ready‑to‑Use Models

Ollama is a platform that makes it simple to download and run open‑source AI models on your local machine.

You’ll find almost everything you need:

Embeddings for semantic search and knowledge bases

TTS (Text‑to‑Speech) to give your AI a voice

STT (Speech‑to‑Text) to let your AI understand you

A wide range of state‑of‑the‑art models, ready to install and use

They may not always be the absolute best in the world, but they’re among the most reliable and accessible open‑source models you can get today — and Ollama makes installation quick and easy.

👉 Discover Ollama

OpenWebUI: The Leading Open‑Source Interface for Local AI

OpenWebUI is a powerful, open‑source interface that brings a ChatGPT‑style experience to your own machine. It runs locally, is easy to install, and works both offline or connected to the cloud.

With OpenWebUI, you can:

Choose and switch models according to your needs

Customize prompts and workflows for a more personal experience

Add local features like speech‑to‑text, text‑to‑speech, or vector search

Enjoy a clean, user‑friendly interface that makes working with AI feel natural and human

Many systems like n8n excel at creating workflows, but they don’t provide a truly engaging chat interface. That’s where OpenWebUI shines: it gives you the comfort and interactivity you need, while staying fully compatible with tools like n8n — so you can build workflows in n8n and use them seamlessly with your local AI in OpenWebUI.

👉 Discover OpenWebUI