[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:personal-ai-router":3},"\u003Ch1>NVIDIA Personal AI Router (PAIR)\u003C\u002Fh1>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fpersonal-ai-router\u002Fblob\u002FHEAD\u002FLICENSE\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-Apache--2.0-blue.svg\" alt=\"License\" \u002F>\u003C\u002Fa>\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fpersonal-ai-router\u002Fblob\u002FHEAD\u002FSECURITY.md\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fsecurity-policy-green.svg\" alt=\"Security Policy\" \u002F>\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>NVIDIA Personal AI Router (PAIR) is a local inference router for a group of\ncompatible computers on the same network. It discovers participating nodes,\nmanages supported inference engines, and presents local proxy endpoints for\nOllama-compatible, OpenAI-compatible, and Anthropic Messages API requests.\nIndependent requests can be routed to eligible nodes according to engine\navailability, model availability, and current workload.\u003C\u002Fp>\n\u003Cp>PAIR is useful for concurrent local workloads such as multi-agent applications.\nPrompts and responses are intended to remain on the local network when every\nconfigured client, model source, engine, and node is local.\u003C\u002Fp>\n\u003Cblockquote>\n\u003Cp>PAIR routes each independent request to one node. It does \u003Cstrong>not\u003C\u002Fstrong> pool GPU\nmemory, combine GPUs into a larger logical GPU, shard one model across\nmachines, or split an in-flight inference request between nodes.\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FNVIDIA\u002Fpersonal-ai-router\u002FHEAD\u002Fassets\u002Fpair-demo.gif\" alt=\"Two paired machines in PAIR's Overview. Requests arrive on one and are routed\nacross both, with each node reporting live GPU and memory use.\" \u002F>\u003C\u002Fp>\n\u003Cp>\u003Cem>Two paired machines: requests arrive on one, run on whichever node suits each\none, and both report live GPU and memory use throughout.\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fpersonal-ai-router\u002Fblob\u002FHEAD\u002Fassets\u002Fpair-demo.mp4\" rel=\"nofollow ugc noopener\">Watch the full clip\u003C\u002Fa>.\u003C\u002Fem>\u003C\u002Fp>\n\u003Ch2>What is supported\u003C\u002Fh2>\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>\u003C\u002Fth>\n\u003Cth>\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd>\u003Cstrong>Operating systems\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Windows 11; Linux; macOS\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Architectures\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>x64 and arm64 on all three. Windows on ARM is experimental.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Installers\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Windows \u003Ccode>.exe\u003C\u002Fcode>; Linux \u003Ccode>.deb\u003C\u002Fcode>; macOS \u003Ccode>.dmg\u003C\u002Fcode>. On other Linux distributions, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fpersonal-ai-router\u002Fblob\u002FHEAD\u002Fdocs\u002Fbuilding.mdx\" rel=\"nofollow ugc noopener\">build from source\u003C\u002Fa>.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Mixing nodes\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Windows, Linux, and macOS nodes can all be paired with each other\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cstrong>Inference engines\u003C\u002Fstrong>\u003C\u002Ftd>\n\u003Ctd>Ollama and LM Studio\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\n\u003Cp>\u003Cstrong>PAIR running on a machine does not mean an engine will.\u003C\u002Fstrong> PAIR itself runs on\nany supported Windows, Linux, or macOS machine. Each engine sets its own requirements\nfor the operating system, GPU, and drivers, and each model needs enough memory to\nload. Whether a particular engine and model work on a particular machine is\nbetween that engine and that machine, so check the engine's own documentation\nbefore assuming a node can serve a model. A node only becomes a candidate for a\nrequest once it is actually running a compatible engine, and PAIR prefers the\nnodes it already knows hold the model.\u003C\u002Fp>\n\u003Ch2>Quick start\u003C\u002Fh2>\n\u003Cp>Download a released build and use the desktop application. That is the path we\nrecommend and the one the rest of this guide assumes. Building from source and\nthe terminal interface both exist for good reasons — changing PAIR, and machines\nwith no desktop — but neither is the ordinary way in. Those are covered in\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fpersonal-ai-router\u002Fblob\u002FHEAD\u002Fdocs\u002Fbuilding.mdx\" rel=\"nofollow ugc noopener\">Building and running PAIR from source\u003C\u002Fa> and\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fpersonal-ai-router\u002Fblob\u002FHEAD\u002Fdocs\u002Fterminal-interface.mdx\" rel=\"nofollow ugc noopener\">Terminal interface\u003C\u002Fa>.\u003C\u002Fp>\n\u003Ch3>Download a release\u003C\u002Fh3>\n\u003Cp>A released installer is signed, sets up the background services and the desktop\napplication together, and adds the firewall rules PAIR needs on Windows. It also\ntells you when a newer release exists and installs it on your say-so from\n\u003Cstrong>Settings → Service\u003C\u002Fstrong>. A build you make yourself is unsigned and checks no\nupdate feed, so you would upgrade it by pulling and rebuilding.\u003C\u002Fp>\n\u003Cp>Download PAIR from the\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FPersonal-AI-Router\u002Freleases\" rel=\"nofollow ugc noopener\">GitHub releases page\u003C\u002Fa>.\nRelease downloads include:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>a Windows installer;\u003C\u002Fli>\n\u003Cli>a Debian package for Linux; and\u003C\u002Fli>\n\u003Cli>a macOS disk image.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cstrong>On Windows and macOS,\u003C\u002Fstrong> double-click the download and follow the installer's\nusual prompts — on macOS that means dragging NVIDIA Personal AI Router to your\n\u003Cstrong>Applications\u003C\u002Fstrong> folder.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>On Linux,\u003C\u002Fstrong> install the package from the directory you downloaded it into:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-bash\">sudo apt install .\u002FNVPAIR-Setup-*.deb\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Cp>If you ha\u003C\u002Fp>\n",1790587592718]