Nvidia PAIR lets you put your idle Macs and PCs to work for AI agents

Nvidia introduced Nvidia Personal AI Router (PAIR), an open-source software router for home networks, on Thursday. This tool lets users utilize idle Macs and PCs within their residence to run small AI models on-demand, accelerating agentic workflows with subagents. Paired with models like NemoClaw, OpenClaw, or Hermes, PAIR allows agents to split tasks among subagents, with model requests running on idle machines.

The software functions as a "virtual inference router" and is built on top of existing Ollama or LM Studio installs on machines. PAIR identifies compatible machines on the local network, using mDNS, and checks their ability to run the requested model. The caveat is that PAIR does not split a single inference request across different machines; instead, it handles each request from start to finish on a single node.

PAIR supports Windows, macOS, and Linux machines with compatible GPUs, such as Nvidia GeForce RTX 20 series and newer, Macs with M4 silicon or newer, or Nvidia DGX Spark. Each system can host different models, but PAIR will only route requests to machines with the required engine and available model.

By monitoring computer availability, PAIR stops the local inference engine once the user resumes work or gaming. In an example, using PAIR with two PCs sporting high-end RTX 5090 GPUs and Qwen3.6 35B A3B model sped up work with five subagents by approximately 1.6x. While few possess multiple RTX 5090 cards, a combination of Mac Studio, Mac minis, and gaming PCs could still improve local agent workflows.

PAIR is currently available as a beta. To join the network, users must install PAIR on each machine, discover and pair the systems, and ensure Ollama or LM Studio is installed, along with downloaded models. PAIR can aid in setup by installing Ollama or LM Studio and initiating model downloads on paired machines.