NVIDIA PAIR 2026: Open Source Virtual AI Router Released
NVIDIA releases Personal AI Router (PAIR) v0.1.1, an open-source virtual inference router distributing local AI requests across RTX, DGX Spark, and Mac nodes.
Stock photo for illustration only, not from the actual event
- NVIDIA announced PAIR public beta (v0.1.1) as an open-source project under the Apache 2.0 license on GitHub.
- Acts as a virtual inference router to schedule and distribute local AI requests across home network nodes.
- Seamlessly integrates with Ollama and LM Studio while exposing OpenAI-compatible proxy endpoints.
- Demonstrated significant speedups, reducing synthetic workload times from 18 minutes down to 8 minutes and 48 seconds.
NVIDIA has announced the release of Personal AI Router (PAIR) version 0.1.1 this week, functioning as a virtual inference router. Rather than serving as a brand-new inference engine, the system discovers compatible hardware on a home network and schedules independent inference requests across them, utilizing existing execution engines like Ollama or LM Studio.
PAIR is fully deployable today, shipping as a public beta with signed installers for Windows, macOS, and Linux, alongside its complete source code available on GitHub under the Apache 2.0 license. The system runs entirely within the local network, requiring internet access solely for downloading models.
A key design decision behind PAIR is that it introduces no new cluster API. Instead, it proxies the Ollama-compatible and LM Studio-compatible interfaces that agent harnesses already use, taking over their default ports. Developers can also configure proxy ports or utilize exposed OpenAI-compatible endpoints, ensuring existing agent harnesses require zero modifications.
Stock photo for illustration only, not from the actual event
For automatic discovery, PAIR leverages mDNS to find nearby systems, allowing manual IP address entry if discovery fails. Trust is established using a six-digit PIN displayed on the inviting machine and entered on the remote node, securing all further communications via generated mTLS certificates.
Each participating node operates Ollama or LM Studio, with PAIR capable of installing engines and initiating model downloads across paired systems to minimize setup friction. A node becomes eligible for requests only when the required engine is active and the exact target model is loaded locally.
"On one RTX Spark laptop, the workload took 18 minutes on average. On a three-device PAIR cluster consisting of an RTX Spark laptop, a DGX Spark, and an RTX 5090, it averaged 8 minutes and 48 seconds."
NVIDIA Demonstration
The introduction of NVIDIA's Personal AI Router addresses critical hardware bottlenecks in local LLM deployments by orchestrating distributed workloads across heterogeneous consumer and workstation hardware. By treating existing local instances as an elastic pool without requiring complex cluster APIs, PAIR democratizes multi-device AI execution for developers and researchers.
PAIR supports GeForce RTX 20 Series GPUs and newer, RTX PRO workstation GPUs starting from Turing, DGX Spark, and Apple M4 silicon or newer. Nodes across Windows, Linux, and macOS can be interconnected freely, with further documentation and resources available via the official NVIDIA technical blog and GitHub repository.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment