The project
Existing dictation tools share one flaw: they take you elsewhere. Open a window, speak, copy, go back, paste. Oratio removes the trip. Press a global hotkey, an overlay appears above any application, you speak, and the text lands in the field you already had focused.
Transcription happens entirely on the machine: whisper.cpp runs on the GPU through Vulkan, so it works on NVIDIA, AMD and Intel Arc cards without vendor lock-in. Audio never leaves the computer, and no cloud subscription is needed for recognition. Voice activity detection ends the recording on its own when you stop talking.
AI rewriting is optional and comes in modes: clean text, email, notes, task list, prompt. The mode can be switched after you have spoken, and the transcription is reprocessed without repeating it. It is the only step that needs the network, and the API key lives in the Windows credential manager, never in a plain file.
The whole thing is built on Tauri: the interface is React, but the core is Rust and the shipped binary is a fraction of what the same app would weigh in Electron. Global hotkeys, the tray icon and per-screen window resizing all go through the native side.
