Flash one USB. Boot any old Optiplex. Get a private AI box on your network —
cloud-key fallback (Anthropic / OpenAI / Gemini) plus the upcoming Polaris Core
local model. Configure via your phone's browser at polaris.local:8080.
# Linux/macOS sudo dd if=polaris-os-v0.0.1.img of=/dev/sdX bs=4M status=progress sync # Then plug the USB into any Optiplex, boot, and visit: http://polaris.local:8080
A custom QLoRA fine-tune of Qwen 2.5 7B trained on the maintainer's repos, memory, and Polaris docs — ships as a ~4.5 GB GGUF via Ollama. Training data is built (~15k examples, 42% custom voice). One GPU rental away from a usable model.