Setting Up a Local AI Cluster at Home with NVIDIA PAIR on Omarchy Linux

Summary: It explains how, when a single computer's VRAM is not enough for large models, several machines on the local network can be used as a single cluster with NVIDIA PAIR. The article gives step by step how PAIR works, its installation on Omarchy, pairing the devices, using it together with Ollama and LM Studio, and a window rule for Hyprland.

While the parameter sizes of artificial intelligence models are growing rapidly and multi-agent workflows have become my daily software development routine, being stuck with a single computer's GPU memory (VRAM) was one of the biggest bottlenecks.

NVIDIA's new open-source tool NVIDIA PAIR (Personal AI Router), announced in September 2026, targets exactly this problem: by connecting all compatible computers on your local network (LAN) (NVIDIA GeForce RTX 20 series and above, RTX Workstation, DGX and even Apple M-series Macs), it lets you create a personal, distributed artificial intelligence inference cluster (Personal AI Cluster).

In this article, I explain step by step how to install NVIDIA PAIR on our Arch Linux-based modern operating system Omarchy 4.0.2 (Hyprland), how to pair it with other devices on the local network (pairing), and how to distribute multi-agent workloads with Ollama / LM Studio.

What Is NVIDIA PAIR and How Does It Work?

NVIDIA PAIR is not a hardware device; it is a lightweight, smart inference router (Virtual Inference Router) running on your local network.

Personal Artificial Intelligence Cluster (AI Cluster) Topology: Zero Cloud Dependency
Client and Main Station: Omarchy Linux (Client / Main PC) · Agentic AI / IDE / Development Environment

| (localhost:8080/v1)

Central Virtual Router: Dynamic Load Distribution, Automatic Idle GPU Detection and Smart Routing (NVIDIA PAIR (Personal AI Router))
NODE 1: 8GB GDDR6 VRAM, low-latency local models. (Mobile Inference, Game Garaj (RTX 4060), Ollama: Qwen 2.5 Coder / Llama 3.2)
NODE 2: RTX 3090 (24GB VRAM) / Apple M-Series Unified Memory. LM Studio: DeepSeek V3 / Heavy Reasoning Models (Server Inference, Homelab Server / Mac)

Key Features:

Dynamic Load Distribution: When an agent sends a query to the main model, it routes subtasks (code analysis, web search summarization, etc.) to other idle GPUs on the network.
Flexible and Seamless: When gaming or rendering starts on a computer on the network, PAIR detects that device as busy and instantly moves the load to the main machine or other idle nodes.
Zero Cloud Dependency (Zero-Trust and Privacy): All prompts, code and model weights travel only over your home's local network (LAN).
Cross-Platform: It offers seamless clustering support across Linux, Windows 11 and macOS.

1. Installation on Omarchy Linux

The cleanest and most practical way to install NVIDIA PAIR on our Omarchy (Arch Linux) system is to integrate the official .deb package into our local file tree.

Open your terminal (Super + Enter) and run the following commands:

# 1. Geçici kurulum dizini oluşturun ve paketi indirin
mkdir -p /tmp/nvpair && cd /tmp/nvpair
curl -LO https://github.com/NVIDIA/Personal-AI-Router/releases/download/v0.1.1/NVPAIR-Setup-0.1.1-amd64.deb

# 2. Paketi açın ve /opt/nvpair dizinine çıkartın
ar x NVPAIR-Setup-0.1.1-amd64.deb
tar -xf data.tar.xz

# 3. İlgili dizinleri sisteme kopyalayın
sudo cp -r opt/* /opt/
sudo cp -r usr/share/applications/* ~/.local/share/applications/ 2>/dev/null || sudo cp -r usr/share/applications/* /usr/share/applications/
sudo cp -r usr/share/icons/* ~/.local/share/icons/ 2>/dev/null || true

# 4. Kolay çalıştırma için global sembolik link tanımlayın
sudo ln -sf /opt/nvpair/nvpair /usr/local/bin/nvpair 2>/dev/null || true

# 5. Temizlik
cd ~ && rm -rf /tmp/nvpair

After the installation is complete, you can launch the application by typing nvpair in the terminal or by searching for NVIDIA PAIR in your Omarchy menu (Super + Space).

2. Pairing Devices (Cluster Pairing)

To fully harness the power of your AI cluster, also install the PAIR application on the second computer on your network (for example, your home desktop PC or homelab server):

Open the PAIR Interface: Run the PAIR application on both devices.
Automatic Discovery: PAIR will automatically list other devices running PAIR on the local network via mDNS.
Pairing Code: Click on the other device, select “Pair”, and verify the confirmation code shown on the screen.
Both devices are now synchronized as a single virtual artificial intelligence pool (Inference Pool)!

3. Ollama and LM Studio Integration

NVIDIA PAIR provides a standard OpenAI-Compatible API Proxy (http://localhost:8080/v1) for your artificial intelligence clients.

# PAIR proxy üzerinden model sorgulama örneği
curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama3.3:8b",
    "messages": [{"role": "user", "content": "Omarchy Linux üzerinde yapay zeka kümesi nasıl optimize edilir?"}]
  }'

PAIR analyzes the GPU memory status on the network, forwards this request to the most suitable node, and returns the response to your main screen within milliseconds.

Recommended Hyprland Window Rule for Omarchy

If you want PAIR to open as a sleek and fixed floating window on Omarchy's dynamic tiling manager Hyprland, you can add the following line to your window rules (~/.config/hypr/hyprland.conf):

windowrulev2 = float, class:^(nvpair|Personal-AI-Router)$
windowrulev2 = size 900 650, class:^(nvpair|Personal-AI-Router)$
windowrulev2 = center, class:^(nvpair|Personal-AI-Router)$

My take

Without spending a fortune on cloud APIs and without sending your personal data to third-party servers, turning all the idle GPUs at home into a single giant artificial intelligence brain is now child's play with NVIDIA PAIR.

Combined with Omarchy's low-resource, fluid Hyprland desktop environment and the raw power offered by our Game Garaj RTX 4060 hardware, managing an at-home artificial intelligence cluster becomes both stable and enjoyable.

Technology, Linux and Open Source Notes — ozanozdil.com