Nvidia PAIR turns idle Macs, PCs into an AI agent cluster

Nvidia has released the Personal AI Router (PAIR) in open beta, a free tool that lets AI agents route inference requests to idle Macs and PCs across a home network instead of running everything on a single machine. Nvidia describes it as a "virtual inference router," not a new inference engine: PAIR sits in front of the Ollama or LM Studio installs already running on each machine, discovers paired systems using mDNS, and sends each request to whichever node has the required engine and the exact model available.

The tool doesn't pool GPUs or split a single inference request across machines. Each paired computer runs a full request start to finish, then hands the response back through PAIR, so an agent still sees one connection while PAIR handles placement behind it. When someone reclaims a machine for other work, the local inference engine on that machine stops automatically.

PAIR targets multi-agent workflows, where a lead agent splits a task across several subagents and those subagents' model calls can spread across whatever hardware is sitting idle in the house. In Nvidia's own demo, two PCs running RTX 5090 GPUs and the Qwen3.6 35B A3B model saw roughly a 1.6x speedup across five subagents.

Supported hardware spans Windows, macOS, and Linux systems: Nvidia GeForce RTX 20-series GPUs or newer, Macs with M4 silicon or later, and Nvidia DGX Spark, with RTX Spark PCs and laptops added once they ship later this year. PAIR is available now as a beta. Setup can also install Ollama or LM Studio and start model downloads on paired machines automatically.