PAIR tool from Nvidia lets you run AI agents on a personal AI data centre

Nvidia has launched Personal AI Router, or PAIR, an open-source software tool that allows users to connect multiple computers on a home network and use their combined computing power for local AI tasks. The tool is designed to use PCs when they are idle, allowing AI workloads to run across several machines instead of relying on a single GPU.

Nvidia says PAIR can discover compatible computers on a local network, connect them and distribute AI inference tasks between them. The system is aimed particularly at agentic workloads, where a complex task can be broken into smaller jobs that can run in parallel. This can allow multiple PCs to process different parts of a task at the same time.

How Nvidia PAIR works

PAIR does not refer to a hardware router. It is software that creates a network of compatible computers and manages how local AI workloads are distributed between them.

The tool is designed to use computers when they are not being actively used, so that AI workloads do not interfere with regular tasks. Nvidia says PAIR can also detect when a device becomes busy and move work to another computer on the network. For example, a workload running on a desktop can be shifted elsewhere if the user starts playing a game on that PC.

This approach is intended to avoid a situation where an AI task is limited by the computing capacity of a single machine. Instead, multiple systems can work on separate parts of a workload in parallel.

Nvidia’s blog gives the example of a user asking a local AI agent to organise a “Sunday Reset” plan. The task could be divided across several computers, with different machines handling separate parts of the request while the user’s main computer remains available for other work.

Which computers can use PAIR?

Nvidia says PAIR works with systems using GeForce RTX 20-series GPUs and newer, as well as RTX Professional GPUs and DGX Spark systems. Apple Macs with M4 chips or newer are also supported. The software is available for Windows, macOS and Linux.

The tool works with local AI software including Ollama and LM Studio, according to The Verge. Nvidia’s own examples also show PAIR being used with local models and agentic workloads.

The PAIR interface provides a view of connected computers, their available hardware and the workloads being processed. Nvidia’s demonstration shows multiple PCs contributing to AI tasks through the same local network.

Source link