Dataset Visualizer

Browser-based visualizer for multimodal agent recordings. Automatically detects mobile (touch) or desktop (keyboard/mouse) datasets from MCAP files and renders the appropriate overlay on synchronized video playback.

Usage

npm install
npm run dev

Drop your MCAP + Video (+ optional JSON) files into the browser, or load datasets from HuggingFace via URL parameters.

Reference

https://github.com/open-world-agents/owa-dataset-visualizer