Nvidia opened a beta of PAIR, short for Personal AI Router, software that discovers RTX-equipped Windows machines, DGX Spark units and Mac systems already on one network and merges them into a single inference endpoint for AI apps and agents. No extra cabling or rack hardware is required, according to Nvidia.
The pitch pairs capacity with privacy. When several tasks queue up, PAIR is meant to shift requests toward whichever local machine has spare compute, and Nvidia says prompts, files and agent context never leave the home network. At launch it plugs into Ollama and LM Studio instead of requiring a new backend.
Nvidia’s validated hardware list is the load-bearing detail: any GeForce RTX GPU from the 20 series onward, a DGX Spark or GB10 box, or a Mac with an M4 chip or newer, alongside 8GB of RAM and about 20GB of free storage. PAIR runs offline once installed; a connection is only needed to fetch models. It supports Windows 11, DGX OS, Ubuntu and macOS Tahoe, so older machines and unsupported operating systems are excluded.
According to Nvidia’s own product page for Personal AI Router (PAIR), published on nvidia.com.
