Skip to content

PAIR 0.93.0 fork build 2 — vLLM engine, Tailscale peers, headless pairing

Choose a tag to compare

View all tags
@github-actions github-actions released this 04 Sep 13:36
· 23 commits to fork/main since this release
v0.93.0-fork.2
ad5cf77

Fork build of NVIDIA Personal AI Router with three additions proposed upstream (NVIDIA/Personal-AI-Router NVIDIA#9, NVIDIA#10, NVIDIA#11). Use this until they merge, then switch back to NVIDIA's signed releases.

What's in it

  • vLLM as a third inference engine next to Ollama and LM Studio. A vLLM already serving on port 8000 (vllm serve or a container) is adopted in place, including the head node of a multi-node tensor-parallel instance.
  • Nodes across Tailscale or any VPN. Add a machine by address or MagicDNS name and it becomes a full peer: engines, models, telemetry, and inference routed over mutual TLS. No multicast needed.
  • Scripted pairing for headless machines: NVPAIR_PIN=123456 nvpair-tui accept over SSH, plus invite, pending, members.

Install

Guide: INSTALL-FORK.md

Headless Linux or macOS, one command, no Go or Node needed:

curl -fsSL https://raw.githubusercontent.com/cguldogan/Personal-AI-Router/feat/vllm-tailscale/scripts/install-headless.sh | bash

Desktop: NVPAIR-Setup-*.dmg (macOS Apple Silicon), .deb (Linux x64 and arm64), .exe (Windows x64). Unsigned: on macOS right-click → Open the first time; on Windows accept the SmartScreen prompt. These packages check no update feed; upgrade by installing the next release.

pair-services-* archives are the bare service binaries for headless use on each OS and CPU; the installer above downloads the right one.

Versions

Services 0.93.0 (broker 0.41.0, OpenAI-compatible proxy 1.0.0, engine manager 0.18.0, terminal interface 0.8.0). The desktop package file names carry the desktop app's own version, 0.1.1.

Assets 12
Loading