Offline transcription. Audio is transcribed locally with Whisper. Nothing is sent to a server. On devices with WebGPU (including recent iPhones) the engine runs on the GPU and is much faster; otherwise it falls back to the CPU automatically. The badge at the top right shows which backend is active. The first run downloads the chosen model once, then it is cached. After transcribing, each line in the transcript is a clickable timestamp that jumps the playback to that moment, and the current line highlights as the audio plays. The microphone needs a secure context, so open this over https or localhost; on iOS it must be Safari itself, not a home-screen app or in-app browser. The engine is fetched from a CDN at first use, so the very first transcription needs a network connection.