Hex: local voice dictation that pastes into the focused desktop app

Typing a long prompt, code comment or quick note one character at a time can break the flow. Hex is an open-source desktop dictation app from anomalyco/hex (Rust, MIT; 5,852 GitHub stars as of November 9, 2026): hold a shortcut to speak, release it, and Hex transcribes on the device and pastes into the focused app. The first run downloads a speech model; the project says transcription then stays on-device instead of sending the recording away.

Core features

• Hold, speak, release: the shortcut starts recording; release transcribes and inserts the result in the current app, with no separate dictation window or copy-and-paste step.
• On-device recognition: the first run needs an internet connection to download a speech model; later transcription runs locally. The README does not state model size or list supported languages, so storage needs and Chinese recognition quality should be checked in practice.
• Three desktop platforms: macOS has an Apple-silicon build; Windows offers x64 and ARM64 installers; Linux has x86_64 and ARM64 builds. The Windows and Linux setup guides label those ports beta.
• Stable and Dev channels: the README lists Stable as the current release; Dev tracks the newest dev branch build and updates itself from the Dev channel. Start with Stable for daily use.
• MIT licensed: the implementation can be inspected or built from source; the project says it does not take pull requests and directs improvement proposals to issues.

Typical use cases

• Long prompts: dictate a detailed instruction into a chat box or code editor and keep the thought intact as the text lands at the cursor.
• Quick notes and messages: speak a short idea into the email, document or task field already open, without switching to a recording app.
• People who prefer local processing: after the one-time model download, the project says recognition stays on-device — useful when that matters and the initial download is acceptable.

Quick start

On an Apple-silicon Mac, install with the Homebrew cask command from the README:

brew install --cask anomalyco/tap/hex

The first launch downloads a speech model. Then hold the configured shortcut while speaking and release it to paste the transcript into the focused app. On Windows, Ctrl+Space is the default in Hold mode; release Space to finish. Windows and Linux users should read their platform guide first for beta status, processor and desktop-environment requirements.

Local dictation boundaries

Hex focuses on turning speech into text in the current app; the README does not describe meeting notes or text organization. For live meeting captions, translation and notes, see Cicada; to turn existing audio or video into structured notes, see AudioNotes.

Platform support has distinct limits: the macOS download is Apple-silicon only; Windows x64 needs an AVX2-capable processor; the Linux beta targets X11 and some wlroots Wayland desktops, and the window still needs a Vulkan driver even when transcription is set to CPU. On Windows, the default is Ctrl+Space; Hex consumes the chord, so an input method or app using it may need a new binding. On Linux, paste uses wtype on compatible wlroots compositors and the RemoteDesktop portal elsewhere; shortcuts use the GlobalShortcuts portal when available or desktop keybindings otherwise.

Summary

Hex suits people who dictate short text into chats, editors and documents and want recognition to stay on their device; teams needing meeting summaries, speaker labels or batch file transcription should use a fuller transcription tool. It is an MIT-licensed Rust desktop app, with a first-run model download, Apple-silicon-only macOS support and Windows/Linux ports still marked beta. One community caveat: the project does not accept pull requests and asks contributors to open issues instead. Check the shortcut and paste behavior on your own desktop before relying on it every day.