Releases: pHequals7/muesli
Muesli 0.5.5
Muesli 0.5.4
Muesli 0.5.4
Native macOS app — dictation + meeting transcription on Apple Silicon.
Install
- Download
Muesli-0.5.4.dmg - Open the DMG and drag Muesli to Applications
- Launch from Applications
Signed, notarized, and stapled by Apple.
Muesli 0.5.3
Muesli 0.5.3
Native macOS app — dictation + meeting transcription on Apple Silicon.
Install
- Download
Muesli-0.5.3.dmg - Open the DMG and drag Muesli to Applications
- Launch from Applications
Signed, notarized, and stapled by Apple.
Muesli 0.5.2
Muesli 0.5.2
Native macOS app — dictation + meeting transcription on Apple Silicon.
Install
- Download
Muesli-0.5.2.dmg - Open the DMG and drag Muesli to Applications
- Launch from Applications
Signed, notarized, and stapled by Apple.
Muesli 0.5.1
Muesli 0.5.1
Quality-of-life improvements and bug fixes.
What's New
- Unified floating indicator — hold-to-dictate now uses the same thin pill as handsfree mode with cancel (X) and stop buttons
- Hidden indicator mode — toggle off the floating mic in Settings; it still appears during active dictation and vanishes when done
- ChatGPT sign-in improvements — OpenAI logo on button, spinner during OAuth, green "Signed in" state, error feedback on failure
- Delete confirmation — dictation delete now asks for confirmation before removing
- Anonymous telemetry — opt-in usage analytics via TelemetryDeck (no content or PII transmitted)
Security
- ChatGPT OAuth tokens moved from Keychain to encrypted file storage (fixes unsigned build issues)
- Token file restricted to owner-only permissions (0600) and excluded from backups
- OAuth state parameter validated to prevent CSRF attacks
Install
- Download
Muesli-0.5.1.dmg - Open the DMG and drag Muesli to Applications
- Launch from Applications
Signed, notarized, and stapled by Apple.
Muesli 0.5.0
Muesli 0.5.0
The first signed, notarized, and stapled public release. No more Gatekeeper warnings — just download, drag to Applications, and launch.
What's New Since 0.4.0
ChatGPT OAuth Sign-In
- Sign in with your ChatGPT Plus/Pro subscription for meeting summarization — no API key needed
- Browser-based OAuth with PKCE, tokens stored securely in macOS Keychain
- Model selection: GPT-5.2, GPT-4.1 Mini/Nano, GPT-4o
Speaker Diarization
- Meeting transcripts now identify individual remote speakers (Speaker 1, Speaker 2, etc.) instead of generic "Others"
- Powered by FluidAudio's pyannote-based CoreML diarization model running on Apple Neural Engine
- Automatic speaker label assignment in first-appearance order
VAD-Driven Chunk Rotation
- Meeting mic audio now splits at natural speech boundaries using Silero VAD streaming detection
- Replaces fixed 30-second chunks — no more mid-sentence cuts
- 3s minimum / 60s maximum chunk duration safety rails
Camera-Based Meeting Detection
- Webcam turning on instantly triggers meeting detection notification
- Event-driven via CoreMediaIO — no polling, zero latency
- Strongest detection signal: camera on = meeting, regardless of which app
New ASR Models
- Qwen3 ASR — 52 languages, ~900MB (int8), 2-3s latency. Warmup inference during preload eliminates 30s first-run penalty.
- Nemotron Streaming (Experimental) — NVIDIA RNNT streaming model. English-only, best for hands-free mode with 10s+ utterances. ~0.3s warm latency. Live text-at-cursor in handsfree mode.
Dictation History
- Infinite scroll — see all your dictations, not just the last 50
- Dynamic date filtering: Last 2 days, Last week, Last month, etc.
- Filter options appear based on your actual data range
Build & Distribution Fixes
- Sparkle.framework properly bundled and deep-signed (all nested binaries)
- App notarization working — signed, notarized, and stapled by Apple
- Status bar only shows downloaded models (not all available)
Install
- Download
Muesli-0.5.0.dmg - Open the DMG and drag Muesli to Applications
- Launch from Applications — no Gatekeeper warnings
Signed, notarized, and stapled by Apple.
Requirements
- macOS 14.2+ (macOS 15+ for Qwen3 ASR and Nemotron Streaming)
- Apple Silicon (M1/M2/M3/M4)
Models
| Model | Size | Languages | Latency |
|---|---|---|---|
| Parakeet v3 (recommended) | ~250 MB | 25 | ~0.13s |
| Parakeet v2 | ~250 MB | English | ~0.13s |
| Qwen3 ASR | ~900 MB | 52 | ~2-3s |
| Whisper Small | ~190 MB | English | ~1-2s |
| Whisper Medium | ~1.5 GB | English | ~2-3s |
| Whisper Large Turbo | ~600 MB | Multilingual | ~2-4s |
| Nemotron Streaming | ~600 MB | English | ~0.3s |
Models download on first use from HuggingFace (~250MB for default Parakeet v3).
Muesli 0.4.0 (Pre-release)
Muesli 0.4.0
Native macOS app — dictation + meeting transcription on Apple Silicon.
What's included
- Push-to-talk dictation with on-device STT (Parakeet, Whisper)
- Meeting recording with mic + system audio capture
- AI-powered meeting summaries (OpenAI, OpenRouter, or ChatGPT subscription)
- ChatGPT OAuth sign-in — use your existing Plus/Pro subscription
- Sparkle OTA update framework
Install
- Download
Muesli-0.4.0.dmg - Open the DMG and drag Muesli to Applications
- Launch from Applications
Note: This build is signed with Developer ID but not notarized. On first launch, right-click > Open or allow via System Settings > Privacy & Security.