LazySketch turns a live transcript into a presentation as you speak. A fast LLM follows the talk and continuously draws hand-sketched diagrams, bullets, code blocks, and the occasional rubber-duck stamp onto a lamp-lit paper canvas.
Features
- Voice-driven slides — the deck follows the live transcript instead of a prepared sequence
- Local speech models — Whisper and Kokoro run in browser workers, with WebGPU acceleration when available
- Bring your own model — connects to any OpenAI-compatible endpoint, without a bundled service or account
- Browser-local storage — presentations and recordings stay in IndexedDB and can be exported or imported
- Cue files — a talk outline gives the drawing agent context without scripting every slide
- Replay and rehearsal — rehearse from a prepared script, record live sessions, scrub through them, or resume for questions
- Agent-friendly CLI — author sheets and cues as files, sync them into the browser, run scripted sessions, and inspect evaluation reports