Muesli Windows

The native Windows app for Muesli — a local-first, privacy-focused dictation and meeting transcription tool.

What is Muesli?

Muesli runs entirely on your laptop. No cloud transcription. No data leaves your machine.

• Dictate anywhere with a global hotkey (F8 by default). Speak, release, and your transcript is pasted into the active app.
• Record meetings with automatic detection of Google Meet, Zoom, Microsoft Teams, and Webex.
• Transcribe locally using open-source Whisper models (CPU or NVIDIA GPU).
• Clean up transcripts with optional local Qwen post-processing.
• Keep everything in a searchable, persistent history.

Privacy

Muesli is built around local-only processing. Concretely:

• Transcription is on-device. Bundled Whisper models run inside a CPython 3.12 subprocess that ships with Muesli. Your audio is never uploaded.
• Microphone is only active during dictation or meeting recording. While you hold the dictation hotkey, or while a meeting recording is running, the WASAPI capture stream is open. At every other moment it is closed.
• System audio loopback is opt-in per meeting. Loopback (the "other side" of a Zoom / Teams / Meet call) only starts after you explicitly click Join & Record on the meeting-detection toast. No silent background capture.
• The global keyboard hook only inspects key codes. Muesli installs a WH_KEYBOARD_LL hook so the dictation hotkey works in any app. The hook checks whether the pressed key matches your configured shortcut and nothing else — no keystroke content is logged, transmitted, or stored anywhere.
• Crash reporting is opt-in and off by default. If you toggle it on (during onboarding or in About → Privacy), stack traces and the app version are sent to Sentry; transcripts, recordings, file paths, and your Windows username are scrubbed before send.
• Local storage paths. Settings and history live in %APPDATA%\muesli\. Whisper / Qwen / Parakeet / pyannote model weights cache to %USERPROFILE%\.cache\muesli\. Captured audio files live in %APPDATA%\muesli\captures\. Nothing leaves these locations unless you ask it to.
• Cloud meeting summaries require your own keys. OpenAI / OpenRouter summary providers are only used when you enter your own API key in Settings. The default summary provider runs locally.

Features

• Dictation — Hold F8, speak, release. Auto-paste or clipboard.
• Local Transcription — Whisper runs on-device. No internet needed.
• Meeting Recording — Auto-detects active meeting windows, captures mic + system audio.
• Meeting Summaries — Auto-generated notes via local rules, OpenAI, or OpenRouter.
• Custom Dictionary — Phrase → replacement pairs for consistent terminology.
• Organized History — Dictations and meetings with folders, search, filtering.
• Dark & Light Theme — Native WPF theming.
• Floating Indicator — Compact pill showing recording/transcribing state.
• GPU Acceleration — Optional NVIDIA Parakeet fast path.
• Privacy First — All processing local. Cloud APIs only if you add your own keys. Crash reporting is opt-in.

Installation

Download Muesli-win-Setup.exe from the latest release and run it. Muesli installs into %LocalAppData%\Muesli and launches itself when it's done.

On first launch, Muesli walks you through onboarding (microphone, hotkey, optional crash reporting) and then downloads the base Whisper model in the background. After that, dictation works offline.

Updates

Muesli checks for updates on every launch and notifies you when one is ready. You can also check manually from About → Check Now, which becomes Restart and update once an update has downloaded. Updates are delivered as Velopack delta packages so subsequent versions are small downloads.

If you installed an earlier v0.2.x build via the Inno Setup .exe, uninstall it first from Settings → Apps before running Muesli-win-Setup.exe. The new installer lives at a different path and won't auto-replace the old one.

Quick Start

1. First Launch — Complete onboarding: pick microphone, hotkey, indicator preference, crash reporting opt-in.
2. Wait for the Base Model — Muesli auto-downloads the recommended Whisper base model after onboarding. A toast tells you when it's ready.
3. Test Your Mic — Models → Test microphone.
4. Dictate — Hold F8, speak, release.
5. Detect Meetings — Open Google Meet, Zoom, Teams, or Webex. Muesli prompts you to record.

Architecture

The WPF app communicates with a Python worker process via JSON-RPC over stdin/stdout. The worker handles model loading, transcription, and optional post-processing. The UI doesn't care whether the backend is Whisper CPU, Whisper CUDA, Parakeet, or a future native engine.

A bundled CPython 3.12 runtime ships inside the installer at <install>\python\, with the base Whisper worker dependencies preinstalled into python\site-packages-muesli\. End users never need to install Python themselves.

Configuration

Settings stored in %APPDATA%\muesli\windows-settings.json:

Setting	Default	Description
Hotkey	F8	Global dictation shortcut
AsrEngine	whisper	whisper or parakeet-v3
ModelProfile	base	Whisper model size
PasteBehavior	active-app	active-app or clipboard
Theme	dark	dark or light
CrashReportingEnabled	false	Opt-in Sentry crash reporting

Optional environment variables:

• OPENAI_API_KEY, OPENROUTER_API_KEY — meeting summary providers (or enter in Settings).
• MUESLI_PYTHON — override the bundled Python with a custom interpreter.
• MUESLI_SKIP_AUTODOWNLOAD=1 — skip the first-launch base-model auto-download.
• MUESLI_SENTRY_DSN — override the embedded Sentry DSN (dev / staging).

Known Limitations

• Not code-signed yet; Windows SmartScreen may warn on first launch.
• System loopback capture can be blocked by some audio setups; mic recording still works.
• Meeting speaker separation is basic (mic = You, loopback = System audio) unless diarization is enabled.
• Meeting detection is heuristic (app titles + browser URLs + process names).

Roadmap

See docs/ROADMAP.md.

For Contributors

Want to build from source, hack on the WPF UI, or work on the Python worker directly? You don't need the installer.

Prerequisites

• Windows 10/11 x64
• .NET 8 SDK
• Python 3.11 or 3.12 on PATH (only for running unpackaged — the installer bundles its own copy)

Build & run from source

.\scripts\setup-worker-runtime.ps1
dotnet run --project .\windows-native\Muesli.Windows\Muesli.Windows.csproj

Optional add-ons (install extra worker deps into the same .venv):

.\scripts\setup-worker-runtime.ps1 -WithPostProcessing   # Qwen cleanup
.\scripts\setup-worker-runtime.ps1 -WithParakeet         # NVIDIA Parakeet
.\scripts\setup-worker-runtime.ps1 -WithDiarization      # pyannote

Package

.\scripts\package-windows-v1.ps1       # ZIP + Velopack Setup.exe + .nupkgs
.\scripts\test-windows-package.ps1     # Smoke test

Releases are produced by .github/workflows/release-package.yml (manual workflow_dispatch). The workflow embeds secrets.SENTRY_DSN into the Release build and uploads the zip + Velopack artifacts.

Tech stack

• UI: WPF .NET 8, XAML, Inter font
• Audio: NAudio (WASAPI microphone + system loopback)
• Transcription: Python worker — Faster-Whisper, NVIDIA Parakeet (optional), Qwen post-processing
• Crash reporting: Sentry (opt-in)
• Updates: Velopack 1.0.x with GitHub Releases as the source

Contributing

Contributions welcome! Priority areas:

• Visual polish (match OG macOS design system)
• Meeting detection reliability
• GPU path testing on diverse NVIDIA hardware
• Code signing once a certificate is available

Open an issue or PR at github.com/Muesli-HQ/Muesli-Windows.

Acknowledgements

Muesli is developed across native platform apps in the Muesli organization. The Windows app follows the same product direction and design language as the macOS app.

Built with:

• Whisper by OpenAI
• Faster-Whisper
• NAudio
• Velopack
• python-build-standalone
• Inter font family

License

MIT License — free for personal and commercial use.