Nvidia PAIR lets you put your idle Macs and PCs to work for AI agents
The New Stack Frederic Lardinois ● Covered by 6 sources
Nvidia’s new PAIR software uses idle Macs and PCs to run local AI agents. It won’t split one job across machines, but it can speed up parallel subagents.
Based on reporting by The New Stack, Frederic Lardinois — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Nvidia is pushing harder on local AI, and PAIR is the latest proof. The new open source tool, called the Nvidia Personal AI Router, turns spare Macs and PCs on a home network into places where small models can run on demand. It’s aimed at agentic workflows, especially setups that break work into subagents and run those tasks in parallel.
PAIR is not a new inference engine, and Nvidia is careful about that. It sits on top of Ollama or LM Studio installs already running on each machine, discovers systems on the local network through mDNS, and routes requests to a node that can handle them. The agent only sees one connection. Behind the curtain, PAIR picks one eligible machine, sends the request there, and passes the response back.
That also means it does not turn several computers into one giant AI box. It does not pool VRAM, merge GPUs, or split a single inference request across multiple machines. Nvidia says the router just chooses where a request should run, based on engine and model requirements, and then hands the work to that machine from start to finish.
The hardware support is broader than a lot of Nvidia software, which is the interesting part. PAIR works on Windows, macOS, and Linux, and Nvidia’s baseline is GeForce RTX 20 series GPUs or newer, a Mac with M4 silicon or newer, or an Nvidia DGX Spark. RTX Spark PCs and laptops are also on the list once they arrive later this year. That makes the Mac support notable, since those machines have become a popular place to run local models even without Nvidia GPUs.
Nvidia says PAIR can keep track of which machines are free, and if someone returns to a PC to work or game, the local inference engine stops and the GPU is reclaimed. In Nvidia’s own example, two PCs with high-end RTX 5090 GPUs and the Qwen3.6 35B A3B model sped up work with five subagents by about 1.6x. PAIR is in beta now, and setup is fairly straightforward: install it on each machine, let it discover and pair the systems, and make sure Ollama or LM Studio and the needed models are in place. PAIR can even help by installing the engines and starting model downloads on paired machines.
My take — AI-written commentary, not fact-checked reporting
This is the kind of AI feature that makes sense because it’s boring in the right way: use the machines already sitting there instead of asking everyone to buy another box. Nvidia is also quietly admitting that local models and open tooling are where a lot of the real action is. The flashy part isn’t the router; it’s that the company keeps building tools for a world where AI runs close to home, not only in someone else’s data center.
Read more about this at: The New Stack