A Granola and WisprFlow rival that never phones home

What it does

Muesli is a native macOS app that pins real-time dictation and meeting transcription to your Mac’s Apple Silicon chip. Hold a hotkey to dictate text at your cursor with roughly 0.13-second latency, or record a meeting while splitting your microphone from system audio and tagging individual remote speakers. All transcription runs locally via CoreML and the Apple Neural Engine; audio never leaves the machine unless you opt into a cloud provider for meeting summaries.

The interesting bit

The app treats live transcription as an explicit, opt-in engineering choice with two distinct architectures. Nemotron 3.5 offers a single multilingual stream that doubles as the final transcript, while Parakeet Realtime EOU provides a low-latency English preview but delegates the authoritative transcript to a separate meeting model. It also bundles an agent-facing CLI so coding agents like Codex can query meetings and pipe dictations without needing your API keys.

Key highlights

• ~0.13 s dictation latency via Parakeet TDT on the Apple Neural Engine
• Simultaneous mic and system audio capture with VAD-driven chunking at natural speech boundaries
• Speaker diarization for remote callers using a CoreML pyannote-based model
• Camera-based meeting detection for Zoom, Teams, FaceTime, and others
• Google Calendar integration with split “Join & Transcribe” notifications
• Six built-in Apple Shortcuts actions, including Siri voice triggers
• iCloud text sync to an iPhone bridge (audio recordings are never synced)
• Post-meeting hooks that receive a JSON payload on stdin
• Bundled agent CLI exposing stable JSON for meeting and dictation data

Caveats

• Live meeting transcription is disabled by default and requires manually downloading a streaming model
• macOS 14.2+ and Apple Silicon only; Intel Macs are not supported
• The optional Gemma 4 experimental model pulls roughly 2.6 GB of additional weights

Verdict

Mac users who want Granola-style meeting notes or WisprFlow-style dictation without subscription fees or cloud audio should grab it. Anyone on Intel silicon or looking for cross-platform support should look elsewhere.