Voxly

Hold a hotkey, speak, release. Your words are transcribed, cleaned up by AI, and pasted into whatever app you're using.

Quick Start

1. Download the latest release for your platform
2. Run the app
3. Configure your API key in Settings:
   • Groq (free): Get a key at console.groq.com
   • OpenAI (paid): Get a key at platform.openai.com
4. Hold your hotkey (default: Ctrl+Space) and speak
5. Release to transcribe and auto-paste

Optional meeting workflow:

1. Open Settings -> Meetings
2. Acknowledge the recording consent notice
3. Choose screen, microphone, and system-audio capture settings
4. Click Start Recording or use the meeting hotkey (default: Ctrl+Alt+M)
5. Stop the recording, then click Transcribe to create a speaker-labelled transcript with Deepgram

Features

Voice Transcription

Hold your hotkey, speak, release. Voxly records audio from your microphone, sends it to an OpenAI-compatible transcription API, and pastes the result into whatever app has focus. The entire flow happens in a few seconds.

Two hotkey modes are available:

• Hold to talk (default) — hold the hotkey while speaking, release to transcribe
• Press to toggle — press once to start recording, press again to stop and transcribe

Output can be configured to either paste directly (preserving your existing clipboard) or paste and copy the transcription to the clipboard.

Modes

Modes let you run a second LLM call on the transcribed text before it gets pasted. Each mode has a name, a system prompt, and a chat model. The transcription is sent as the user message, and the LLM's response replaces the raw transcription.

Three built-in modes are included:

• Clean Draft — removes filler words and fixes grammar while preserving your tone
• Email Composer — converts spoken draft into a professional email with subject and body
• Developer Mode — formats speech into clear instructions for coding agents (like Claude Code, Cursor, etc.)

You can create custom modes with any system prompt. Activate a mode to use it, or deactivate all modes to paste raw transcriptions. The model list is fetched live from your provider's /models endpoint.

Vocabulary

Define custom word replacements for terms the transcription model frequently gets wrong. Each vocabulary entry maps a word to one or more replacement patterns. When any replacement appears in the transcribed text, it gets corrected to the target word.

Useful for names, technical jargon, or domain-specific terms that speech-to-text models struggle with. Entries can be individually enabled or disabled without deleting them.

History

Every transcription is saved locally with a timestamp. The History tab shows recent transcriptions with relative timestamps (e.g. "5m ago") and exact times on hover. If a mode was active, both the original and formatted versions are shown. You can copy any past transcription to the clipboard or delete individual entries.

Meeting Recording

The Meetings tab records longer calls or screen sessions as local MP4 files. On Windows, Voxly can capture:

• Screen video
• Microphone audio
• System audio from the selected Windows playback device

Meeting recordings are saved under %APPDATA%\dikt\meetings\<meeting-id>\ on Windows. Each meeting keeps recording.mp4 as the source of truth. When both microphone and system audio are captured, Voxly also keeps an aligned two-channel transcript-audio.m4a for more reliable meeting transcription:

• Channel 1: microphone / You
• Channel 2: system audio

You are responsible for getting permission from meeting participants before recording.

Meeting Transcripts

Meeting transcription is manual. After stopping a recording, click Transcribe in the meeting detail view. Voxly extracts or reuses meeting audio, sends it to Deepgram, and saves transcript.json beside the recording.

When separate mic/system channels are available, Voxly uses Deepgram multichannel transcription with diarization so transcript rows can be labelled as You plus distinct remote speakers like Sys-A and Sys-B. If only mixed audio is available, Voxly falls back to speaker diarization with generic speaker labels. Obvious duplicate system-audio bleed picked up by the microphone is filtered from multichannel transcripts.

Floating Overlay

A minimal pill sits at the bottom of your screen showing the current state: recording, transcribing, formatting (when a mode is active), or done. It stays on top of all windows and passes through mouse clicks when not hovered.

System Tray

Voxly minimizes to the system tray. Right-click for quick access to Settings, Reset Position, or Quit. Left-click toggles the overlay visibility.

Supported Providers

Provider	API URL	Pricing
Groq	https://api.groq.com/openai/v1	Free tier available
OpenAI	https://api.openai.com/v1	Pay per use
Custom	Any OpenAI-compatible endpoint	Varies

Voxly uses the OpenAI-compatible API format. Any provider that supports /audio/transcriptions (for speech-to-text) and /chat/completions (for modes) will work with the Custom provider option.

Meeting transcripts use Deepgram separately from the dictation provider. Add a Deepgram API key in Settings -> Meetings to enable meeting transcription.

Platform Support

Voxly is built to be cross-platform (Windows, macOS, Linux), but has only been tested on Windows so far. If you're on macOS or Linux and want to help test, bug reports and feedback are very welcome.

Security & Data Storage

• Short dictation audio is recorded locally and sent to your configured API endpoint for transcription. It is not saved as a reusable recording.
• Meeting recordings are stored locally as MP4 files under the meetings directory until you delete them.
• Meeting transcription audio is stored beside each meeting recording. Clicking Transcribe sends extracted meeting audio to Deepgram.
• Transcription history, meeting metadata, meeting transcripts, and future meeting summaries are stored in a local SQLite database:
  • Windows: %APPDATA%\dikt\dikt.db
  • Linux/macOS: $XDG_CONFIG_HOME/dikt/dikt.db (or ~/.config/dikt/dikt.db)
• Existing transcription_history.json and meetings/index.json data is imported into SQLite automatically on first launch after upgrading. The old JSON files are left in place as a downgrade-safe backup.
• Paste behavior: Voxly writes the transcript to the system clipboard and triggers paste (Ctrl+V on Windows/Linux, Cmd+V on macOS). Clipboard managers may record these changes.
• Settings (provider, base URL, model, hotkey, vocabulary, modes) are stored in a local JSON file:
  • Windows: %APPDATA%\dikt\settings.json
  • Linux/macOS: $XDG_CONFIG_HOME/dikt/settings.json (or ~/.config/dikt/settings.json)
• Meeting files are stored in:
  • Windows: %APPDATA%\dikt\meetings\
  • Linux/macOS: $XDG_CONFIG_HOME/dikt/meetings/ (or ~/.config/dikt/meetings/)
• API keys are stored using the OS credential manager via keyring when available. If unavailable, Voxly falls back to obfuscated values in settings.json.

Development

Prerequisites

• Node.js (v18+)
• Rust
• Platform-specific dependencies:

Windows:

• Visual Studio Build Tools with C++ workload
• FFmpeg available on PATH, bundled with the app, or pointed to by VOXLY_FFMPEG

macOS:

• Xcode Command Line Tools: xcode-select --install

Linux (Ubuntu/Debian):

sudo apt install -y \
  libwebkit2gtk-4.1-dev \
  libgtk-3-dev \
  libayatana-appindicator3-dev \
  librsvg2-dev \
  libasound2-dev \
  libssl-dev \
  libxdo-dev

Run in Development

npm install
npm run tauri dev

Build for Production

npm run tauri build

Tech Stack

• Framework: Tauri 2.x
• Frontend: SolidJS + Tailwind CSS
• Backend: Rust
• Audio: cpal
• Short dictation transcription: OpenAI-compatible Whisper/transcription API (cloud)
• Meeting transcription: Deepgram Nova-3 pre-recorded transcription API (cloud)
• Formatting: OpenAI-compatible Chat Completions API (cloud)

License

MIT