Run any model – LLMs, vision, voice, image, video – on any hardware, no GPU required, with a self-hosted, OpenAI-compatible API.