[Testing] NVIDIA Personal AI Router (PAIR) — unofficial headless container

I’ve created an unofficial container and Unraid template for NVIDIA Personal AI Router, which routes inference requests between paired computers.

The container runs PAIR headlessly alongside a separately installed Ollama backend. Manage pairing and view the node from the NVIDIA PAIR desktop application on another computer.

• Source and documentation

• Docker Hub image

• Image: greggee/nvpair-container:1.0.0

• Linux/amd64 only; not an NVIDIA-supported image.

Requirements

• Ollama running separately and reachable from the Unraid host at 127.0.0.1:11434.

• Host networking and no conflicting PAIR instance on the server.

• A PAIR desktop client, preferably matching version 1.0.0.

• Optional NVIDIA driver/container GPU support for telemetry.

Install the template

Run in the Unraid terminal:

mkdir -p /boot/config/plugins/dockerMan/templates-user

curl -fL \
  https://raw.githubusercontent.com/ggee/nvpair-container/main/unraid/nvpair.xml \
  -o /boot/config/plugins/dockerMan/templates-user/my-nvpair.xml

Open Docker → Add Container, select nvpair, review the settings, and click Apply. Appdata permissions are initialized automatically.

To customize the displayed node name, enable Advanced View and change --hostname=PAIR-UNRAID in Extra Parameters.

Pairing

Send an invitation from your PAIR desktop client, then run:

docker exec --user=10002:10002 -it nvpair python3 /app/admin.py accept

Enter the PIN displayed by the inviting client.

Optional NVIDIA telemetry

Append --gpus=all to Extra Parameters, set NVIDIA GPU selection to all, and leave NVIDIA capabilities as utility. Preserve the other parameters. Ollama’s GPU configuration is separate.

Networking and limitations

Keep TCP 11435, 14318–14323, and UDP 5353 restricted to your trusted LAN and loopback, including IPv6. Do not expose these services to the Internet.

Startup, LAN pairing, and GTX 1080 telemetry have worked on my Unraid system. Broader testing and inference validation are welcome. This wrapper enables only Ollama’s proxy; the desktop still displays LM Studio and llama.cpp install options, which aren’t supported/tested here.