Connect models on a canvas

Drag nodes, wire outputs to inputs, and run multi-step AI pipelines — LLM planning, image/video generation, branching logic.