Build a 100% Client-Side AI Motion Capture Engine in the Browser Imagine uploading a dance video and having it instantly turned into a neon stickman animation—all without uploading a single megabyte to a server. That is stikshot, a local-first, serverless AI motion-capture web app. In this article, we'll dive into the architecture, the browser APIs we used, and the workarounds we engineered to make WebCodecs run stably on desktop and mobile. The Core Stack To keep stikshot completely serverless, static, and scale-free, we built the entire pipeline to run inside the user's browser sandbox: AI Engine: TensorFlow.js running the MoveNet Multipose model. Background Thread: Web Workers to keep the UI responsive. Compression: The browser's native WebCodecs API (VideoEncoder and AudioEncoder). Muxer: webm-muxer to compile the video/audio streams into a playable WebM file.…