Hey everyone 👋 I built Oprel because I kept running into the same problem with local AI tools: They either: Work great on high-end GPUs only Or work everywhere but don’t fully utilize hardware Or crash when memory gets tight Oprel is my attempt to build a smarter local AI runtime. The core idea is simple: Automatically adapt to the user’s hardware. Instead of forcing one backend for everyone, Oprel: Detects GPU / VRAM / RAM Selects the optimal backend Applies quantization automatically Supports hybrid GPU + CPU offloading when VRAM is limited Falls back to CPU cleanly if CUDA fails On mid-range GPUs, it can outperform traditional llama.cpp setups. On low-VRAM GPUs, it can run larger models using hybrid layer splitting. On high-end GPUs, batching increases total throughput significantly. It’s Python-first, installable via pip, and designed for developers who want control without manual tuning. This is still evolving, and I’m actively improving: Stability on Windows Faster model downloads Smarter memory monitoring OpenAI-compatible API Experimental support for image & video generation via ComfyUI integration. If you try it, I’d genuinely love feedback - especially from people running: CPU-only setups 4GB GPUs RTX 3060 / 4070 tier Or high-end batching workloads Thanks for checking it out 🙏

Oprel

Local AI That Actually Uses Your Hardware Properly

Local AI That Actually Uses Your Hardware Properly