Voice-Md is an innovative AI-powered markdown editor that transforms voice commands into dynamic interface interactions. Instead of text-based edits, users speak their intentions, such as “summarize this section,” and the AI agent responds by rewriting content directly within the editor. A key feature is the generative UI layer, A2UI v0.9, which interprets the model’s JSONL output to progressively stream and render React components. This allows for instant variant scrubbing via a length dial and the intelligent presentation of alternative cards, follow-up chips, or no affordance at all, depending on what the model deems most fitting. OpenAI Realtime handles the voice loop, utilizing hover gestures to replace traditional Voice Activity Detection, and Milkdown serves as the underlying editor. This project is a testament to efficient development, working end-to-end with just six dependencies.