Your conversations never leave your device.

muesli's architecture eliminates cloud exposure by design. Here's exactly how.

Data flow

1

Audio capture

System audio captured via macOS APIs. Stays in local memory. Never written to disk as audio.
2

Transcription

Open-source speech-to-text model runs on Apple Silicon. Audio is transcribed on-device. No network requests.
3

Summarization

Open-source summarization model runs on Apple Silicon. Transcript is summarized locally. No network requests.
4

Storage

Notes saved to local SQLite database at ~/Library/Application Support/com.trymuesli.app/. Never synced.

What muesli never does

× Transmit audio to any server
× Send transcripts to cloud APIs
× Connect to third-party AI services (OpenAI, Google, etc.)
× Store data outside your device

On-device models

Model	Purpose	Size	Runs on
Open-source speech-to-text model	Speech-to-text	~1.6 GB	Apple Silicon (local)
Open-source summarization model	Summarization	~1.1 GB	Apple Silicon (local)

Both models download once on first launch and run entirely on-device.

Network activity

muesli makes no network requests during meeting capture, transcription, or summarization. The only network activity is the initial model download on first launch (~2.7 GB total from Hugging Face). After that, muesli works fully offline.