Kolosal AI is an open-source desktop application for running large language models entirely offline on personal devices. Built on the Genta Personal Engine (leveraging llama.cpp), it supports any CPU with AVX2 instructions and AMD/NVIDIA GPUs. The application compiles to approximately 20 MB and targets edge devices like Raspberry Pi, enabling on-premise AI solutions without cloud dependencies. It supports popular open models including Mistral, LLaMA, Qwen, DeepSeek, and Phi variants, and includes capabilities for custom dataset generation and model training alongside inference.