Arion

Arion positions itself as voice input that never leaves your computer. Instead of sending audio to a cloud service, it runs speech recognition on the device itself, so dictation works without a network connection once the model has been downloaded. The project is open source under AGPL-3.0 and offers builds for macOS 14 or later plus a Windows preview.

The core interaction is simple: press Option-Space in any app and speak. Arion recognizes Chinese, English, or a mix of both without requiring you to pick a language first, and the recognized text is typed at the cursor. While you speak, the text stays a draft in a small pill overlay, and only the finished sentence is inserted, which reduces messy partial output.

It is aimed at people who dictate into everyday work apps: chat messages, meeting notes, emails, and discussions that blend technical English with Chinese. Privacy-sensitive users, offline workers, and open-source advocates are the clearest fit, since no audio or text is sent anywhere after setup.

Before relying on it, expect a one-time model download and a permissions onboarding that asks for one permission at a time. The menu bar panel shows the last dictation, and history persists only until you quit. Windows users should treat the current build as a preview. Overall, Arion is a focused, privacy-first dictation tool rather than a full transcription suite.