NVIDIA Releases 'PAIR' Beta for Local AI Inference Integration

NVIDIA has released the beta version of Personal AI Router (PAIR), showcasing its ability to integrate various devices within a home into a single local AI inference cluster. PAIR connects devices such as NVIDIA DGX Spark, Windows systems equipped with RTX, and macOS devices, routing AI app and agent workflows to a single local endpoint.

Integration of Local Computing Resources and Privacy Protection

The core of PAIR is its ability to integrate existing devices on the network into a single personal AI inference cluster within minutes. Without the need for special cables or complex rack setups, the Agent Workload Routing feature intelligently distributes AI inference requests among available local nodes. This maximizes idle computing resources regardless of the OS and ensures data privacy by keeping prompts, files, and agent contexts within the user's home network without being sent to the cloud.

Supported Platforms and Requirements

PAIR is compatible with existing local inference backends such as Ollama and LM Studio, and runs on Windows 11, DGX OS, Ubuntu, and macOS Tahoe. Recommended hardware configurations include GPUs of GeForce RTX 20 series or higher, DGX Spark/GB10, Mac M4 or higher, 8GB or more of RAM, and 20GB or more of disk space. An internet connection is required only for downloading models, while operation itself is possible offline.