https://github.com/ganbarovemin/mywhi
Local-first macOS dictation with WhisperKit
https://github.com/ganbarovemin/mywhi
coreml dictation macos offline-first swift whisperkit
Last synced: 6 days ago
JSON representation
Local-first macOS dictation with WhisperKit
- Host: GitHub
- URL: https://github.com/ganbarovemin/mywhi
- Owner: GanbarovEmin
- License: mit
- Created: 2026-06-21T10:02:35.000Z (17 days ago)
- Default Branch: main
- Last Pushed: 2026-06-21T10:44:12.000Z (16 days ago)
- Last Synced: 2026-06-21T12:19:29.250Z (16 days ago)
- Topics: coreml, dictation, macos, offline-first, swift, whisperkit
- Language: Swift
- Size: 5.94 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# MyWhi
Local-first macOS dictation and meeting transcription app. MyWhi turns spoken Russian or English into clean text for notes, messages, documents, and work calls without sending audio to a cloud transcription service.
It is built as a native Swift / SwiftUI desktop app for macOS with local speech-to-text backends, global hotkeys, Markdown history, and a dedicated Meeting Mode for long calls.
## What MyWhi Does
- **Fast daily dictation**: press `Option` + `Command`, speak, and get text ready to paste or auto-insert into the active app.
- **Voice message transcription**: drag `.wav` or `.m4a` files into MyWhi and convert them to editable text.
- **Meeting Mode**: record work calls, capture microphone and system audio, then prepare transcript, speaker-aware notes, and summary output.
- **Local speech-to-text**: choose between WhisperKit and Soniqo Speech models for offline ASR on your Mac.
- **RU / EN workflow**: Russian, English, and auto language modes for everyday multilingual work.
- **Markdown vault**: save transcripts locally as Markdown in `~/Library/Application Support/MyWhi/vault/`.
- **Native macOS UX**: menu-bar presence, desktop window, global hotkey, Sparkle updates, no Electron.
## Meeting Mode
Meeting Mode is a separate workspace for calls and long-form recordings. It is designed for work meetings where you need the transcript, speakers, and a useful summary after the call.
The current pipeline focuses on:
- microphone recording
- optional system audio capture for video calls
- Soniqo Speech ASR model selection
- noise cleanup toggle
- speaker separation toggle
- context field for names, projects, and domain terms
## Speech Engines
MyWhi supports two local transcription paths:
- **Soniqo Speech**: recommended default for the current daily dictation workflow. The bundled model selector includes Qwen3 0.6B 8-bit, Qwen3 1.7B 8-bit, Parakeet TDT, and Nemotron Streaming options.
- **WhisperKit**: local Core ML / Metal transcription with cached Whisper models for offline use.
## Why Local-First
- Audio stays on this Mac.
- Transcription runs locally through the selected backend.
- MyWhi does not call a cloud transcription API.
- Temporary recordings are stored under `/tmp/mywhi/recordings/`.
- Saved transcript history is plain Markdown.
- Accessibility permission is used only for optional paste/typing behavior.
## Install
Download the latest DMG from [GitHub Releases](https://github.com/GanbarovEmin/MyWhi/releases).
1. Open `MyWhi-3.9.0.dmg`.
2. Drag `MyWhi.app` to `Applications`.
3. Open MyWhi from `/Applications`.
4. Allow Microphone access when macOS asks.
5. Optional: enable Accessibility permission for auto-paste into the active app.
6. Optional: enable Screen Recording/System Audio permission for Meeting Mode call audio.
The current public preview is ad-hoc signed, not Developer ID notarized. macOS Gatekeeper may require right-clicking the app and choosing **Open** on first launch.
## Updates
MyWhi uses Sparkle for app updates. Use either:
- right-click the menu-bar icon and choose **Check for Updates...**
- open **Settings** and click **Check for Updates**
Updates are read from the GitHub-hosted `appcast.xml` in this repository and download release DMGs from GitHub Releases.
## Build From Source
Requirements:
- macOS 14+
- Xcode command line tools
- Swift Package Manager
```bash
git clone https://github.com/GanbarovEmin/MyWhi.git
cd MyWhi
swift test
./build-dmg.sh --install
open /Applications/MyWhi.app
```
Build outputs:
- app bundle: `dist/MyWhi.app`
- DMG: `dist/MyWhi-3.9.0.dmg`
## Release Workflow
```bash
swift test
./build-dmg.sh --install
codesign --verify --verbose=2 dist/MyWhi.app
hdiutil verify dist/MyWhi-3.9.0.dmg
```
Sparkle appcast signing uses an EdDSA private key stored in the local macOS Keychain. The public key is committed in `Info.plist` as `SUPublicEDKey`; the private key must not be committed.
Public distribution still needs Developer ID signing and notarization for a polished Gatekeeper experience.
## Project Layout
```text
Sources/MyWhi/
MyWhiApp.swift App entrypoint and menu-bar integration
AppState.swift Main app state and recording flow
Engine/ WhisperKit and Soniqo transcription backends
Desktop/ Desktop shell, Home, Meeting Mode, Scratchpad, Insights, Settings
Services/UpdateController.swift Sparkle update controller
Resources/
AppIcon.icns
en.lproj, ru.lproj Localized UI strings
docs/
assets/ README screenshots
audits/ UI/UX audit notes
releases/ Release notes
```
## Keywords
macOS dictation, offline speech-to-text, local transcription, WhisperKit, Soniqo Speech, SwiftUI dictation app, meeting transcription, call transcription, Russian ASR, English ASR, voice notes, Markdown transcripts.
## License
MIT. See [LICENSE](LICENSE).