<p>Generative AI has become one of the most important tools, but it usually comes with additional costs, privacy concerns, and vendor lock-in. In this session, Markus Schüttler shares lessons learned from two years of building and maintaining an open-source local AI application in collaboration with Intel - an application that runs entirely on consumer hardware. No cloud, no API keys, no per-token billing.</p> <p>The talk covers the practical challenges of running multiple AI modalities (text, image, video, and voice) on a single device, including shared VRAM management across competing backends, the trade-offs involved in model quantization on consumer hardware, and the evolution of a modular backend architecture designed to support countless open-weight models on Intel and NVIDIA hardware, as well as Apple silicon.</p> <p>In a live demo, Markus Schüttler’s fully local agent will create a browser game, including 2D sprites, 3D assets, and game logic.</p> <p>The session reflects on what works and what does not, and explores where local AI is heading next for developers and organizations that want to own their intelligence rather than rent it.</p>