NVIDIA launches open-source router for local AI workloads
NVIDIA has released Personal AI Router (PAIR), a virtual inference router designed to distribute independent local AI requests across compatible computers on a home network. The public beta, version 0.1.1, includes signed installers for Windows, macOS, and Linux, while its source code is available under the Apache 2.0 license. Internet access is required only for downloading models.
PAIR does not provide a new inference engine or combine hardware into one accelerator. It proxies existing Ollama- and LM Studio-compatible interfaces, allowing agent tools to continue using their current integrations. PAIR uses mDNS for discovery, supports manual IP entry, and requires six-digit PIN pairing before securing traffic with generated mutual TLS certificates.
A request is assigned to one eligible node for its entire run. Eligibility depends on the enabled engine, the exact model being available, current job load, and GPU utilization. In NVIDIA’s Hermes Desktop demonstration, a five-subagent workload using Qwen 3.6 35B A3B averaged 18 minutes on one RTX Spark laptop and 8 minutes 48 seconds on a cluster of that laptop, a DGX Spark, and an RTX 5090. The company describes the figures as unofficial.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.