hehehai/voxt

Voxt is a macOS menu bar app that lets you speak instead of type. Hold a keyboard shortcut, speak, release, and your words are automatically transcribed and pasted into whatever app you were using, a text editor, a chat window, an email draft, anything with a text cursor. Beyond basic dictation, Voxt offers three modes triggered by different shortcut combinations. The first converts speech to text with optional AI cleanup, removing filler words, fixing punctuation, and applying custom rules depending on which app you are using, different rules for coding versus email versus chat, for example. The second mode transcribes speech and then immediately translates the result into another language. The third mode treats your speech as an AI prompt: you say something like "write me a 200 word summary of X" and the response is inserted into your current app. Under the hood, Voxt separates the speech to text step from the language model enhancement step, letting you choose different providers for each. It supports running models entirely on your device, with no data sent to the cloud, including several local speech recognition families and local language models, useful for privacy or offline use. It also connects to external AI provider APIs. A personal dictionary feature lets you add specific terms or names that the transcription engine should recognize correctly, and can auto-correct near matches before output. The app also supports rewriting selected text by voice, such as saying "make this shorter and smoother." It is built natively in Swift and targets macOS, and is available via direct download or Homebrew. The full README is longer than what was shown.