NVIDIA has released the Personal AI Router (PAIR) software in an open-source beta version, which selects available computers within a local network for local AI requests, acting like a scheduler that assigns tasks to idle workstations.

image.png

Automatically discover devices and assign requests, Ollama and LM Studio benefit directly

On compatible devices, PAIR can run on M4 and higher Macs, NVIDIA RTX PCs, and DGX Spark, providing a unified entry point for local AI tools such as Ollama and LM Studio. Users can route multiple independent AI requests to idle devices within a home or office local network.

NVIDIA positions PAIR as a local AI request scheduling software within a local network. It can automatically discover connected computers and assign each request to an available device. The models are still run by Ollama or LM Studio on the target device, so users don't need to configure each machine separately for every AI application. When a device is busy or unavailable, PAIR will redirect new requests to other paired devices. For example, when AI agents review multiple documents simultaneously, they can assign independent tasks to different computers. NVIDIA emphasizes that this mechanism is for distributing concurrent workloads, not implying that a single AI response will automatically become faster by connecting more computers.

Test: Task splitting reduced time from 18 minutes to 8 minutes and 48 seconds

NVIDIA's demonstration showed that using Ollama with the Qwen 3.6 35B A3B model in Hermes, splitting the task among 5 AI sub-agents, the average completion time for a single RTX Spark laptop was 18 minutes; after adding DGX Spark and RTX 5090, the average time was reduced to 8 minutes and 48 seconds.