Koe

About

Koe is a background-first voice input tool for macOS. Press a trigger shortcut (default: Fn, configurable) and speak. You can hold the key, tap once to start and stop, or double-tap to start and single-tap to stop. Audio streams in real time to a cloud ASR service (Doubao/豆包 by ByteDance), and a small floating status pill shows interim recognition text as you talk. When recognition and LLM correction finish, the corrected text is pasted into the active input field. If overlay templates are enabled, you can click a template or press 1-9 to rewrite the result and copy that variant to the clipboard.

Koe keeps its interface minimal. It runs as a menu bar item with the floating pill during sessions, and all configuration lives in plain text files under ~/.koe/, which you can edit with any text editor or through the built-in settings window. The dictionary is a plain .txt file with one term per line. Changes to ASR, LLM, dictionary, and prompt settings apply on the next hotkey press, and hotkey changes are detected within a few seconds, with no restart required.

Koe is built with native macOS technologies, using Objective-C for hotkeys, audio capture, clipboard access, permissions, and paste automation, and Rust for the core. It is Apple Silicon only and installs via Homebrew (brew install koe --cask) or direct download. According to the site, the app stays under 15 MB after installation and typically uses about 20 MB of memory. Updates are delivered in-app through Sparkle.