Nvidia's PAIR routes AI tasks across your home network

Nvidia released Personal AI Router (PAIR) as an open-source beta. It picks which computer on your local network should handle an AI request, like a dispatcher sending jobs to free workers. It works on M4 or newer Macs, Nvidia RTX PCs, and DGX Spark devices. It gives local AI tools like Ollama and LM Studio one shared entry point. You can spread separate AI jobs across idle machines at home or in the office.

PAIR finds computers that have joined and sends each request to one that's free. The model still runs on the target machine through Ollama or LM Studio, so you don't need to set up every app on every computer. If a machine is busy or offline, PAIR sends new requests to other paired devices. For example, when an AI agent reviews several documents at once, it can hand different tasks to different computers.

This is built for spreading out many jobs at the same time, not for making a single AI reply faster. In a demo, using Ollama with Qwen 3.6 35B A3B in Hermes, one RTX Spark laptop took 18 minutes on average for tasks split across 5 AI sub-agents. Adding a DGX Spark and an RTX 5090 cut that to 8 minutes and 48 seconds.