Tools· Meetings & Minutes

Muesli

Open source macOS app for local dictation and bot-free meeting transcription, running speech recognition entirely on Apple Silicon instead of in the cloud.

structuringexecutableOpen Source

Open provider

Description

Muesli is an open source, native macOS app that bundles speech recognition for two use cases: dictation in the flow of everyday work and recording meetings through the computer's own hardware. The app sits in the menu bar, waits for a configurable hotkey, and pastes the recognized text directly at the cursor position in the active application after the hotkey is released, holding the hotkey for short dictations and double tapping it for longer, hands free recordings. For meetings, Muesli captures the microphone and system audio through a CoreAudio process tap, with a ScreenCaptureKit fallback, detects active calls from camera and microphone activity in recognized meeting apps such as Zoom, Teams, or Slack, and can open meeting links straight from calendar events and start the recording.

Speech recognition runs entirely locally on Apple Silicon, through CoreML, the Neural Engine, and Metal, using models such as Parakeet TDT, Nemotron 3.5, Cohere Transcribe 2B, several Whisper sizes through WhisperKit, or experimentally Gemma 4. This requires a Mac with an Apple Silicon processor; the recommended Parakeet v3 model downloads roughly 450 MB on first launch, and larger models such as Gemma 4 several gigabytes. Meeting capture without a bot participant needs microphone permission for your voice, screen or system audio recording permission for computer audio, and optional Calendar access for meeting detection. Pasting the dictated text depends on Accessibility permission, and the global hotkey depends on Input Monitoring; all of these permissions are tied to the specific installation path in /Applications, which means duplicate copies of the app can cause dictation and hotkey permissions to fail silently. Structured notes and summaries built from the raw transcript are an optional second step that requires either your own API key for OpenAI or OpenRouter, signing in with an existing ChatGPT subscription, or a locally running Ollama model.

Compared with dictation tools that send audio to a cloud service for processing, the difference is that Muesli dictates and pastes without an internet connection once installed. Compared with meeting bots that join a video call as their own participant, Muesli instead records from the user's own machine, so no additional guest shows up in the participant list and the raw transcript stays available as a checkable source alongside the generated summary. Anyone looking for pure cloud transcription without a local install, or working on Windows or Linux, will not find that option in Muesli; the app is built exclusively for macOS.

Strengths

Local processing on Apple Silicon

Speech recognition runs through CoreML and the Neural Engine directly on the Mac, at roughly 0.13 seconds of latency via Parakeet TDT on the Neural Engine.

Meeting capture without a bot

Muesli records microphone and system audio through a CoreAudio process tap from the user's own machine, instead of joining the video call as an additional participant.

Multiple interchangeable ASR models

Parakeet TDT and Nemotron 3.5 through the Neural Engine, Cohere Transcribe 2B, several Whisper sizes through WhisperKit, plus Qwen3 ASR, SenseVoice Small, and Indic ASR are available for download in the Models tab.

Speaker detection in system audio

A pyannote based CoreML model from FluidAudio distinguishes individual speakers in the recorded system audio.

Agent friendly CLI

The muesli-cli bundled inside the app delivers transcripts as plain text, JSON, or Markdown, while meeting and dictation lists and individual records come back as JSON.

Assessment

AI features

Automatic meeting notes The transcript becomes structured notes with a summary, decisions, and action items, using OpenAI, OpenRouter, an existing ChatGPT sign in, or a local Ollama model.
Filler word removal "uh," "um," "er," "hmm," and similar verbal disfluencies are automatically stripped from the recognized text.
Personal dictionary with fuzzy matching Custom terms, phrases, and replacement pairs can be stored, and Jaro-Winkler matching automatically corrects transcription errors.
Voice driven Computer Use planner An optional planner can execute local app and browser actions from dictated commands, with a configurable model and timeout.
Post meeting hooks After a meeting finishes, Muesli can trigger a user supplied executable script with the meeting data passed as JSON on stdin.

Suitable for

Individuals on the Mac who want dictation pasted directly into the currently active app, without sending audio to a cloud service
Teams who want meeting transcripts captured without an additional bot participant in the video call
Developers who want to feed transcripts and meeting data into their own scripts or agents through a local CLI
Less suitable for Less suitable for users outside of macOS or without Apple Silicon hardware, since the app is built exclusively for Apple Silicon

Limitations and notes

macOS with Apple Silicon only Muesli is a native macOS app built for Apple Silicon; the vendor does not mention other operating systems.
Permissions tied to the install path Accessibility, Input Monitoring, and recording permissions are bound to one specific app copy in /Applications, so multiple copies or launching from another folder can silently disable dictation and the hotkey.
Live transcription off by default A live preview during a meeting requires downloading a streaming model such as Parakeet Realtime EOU or Nemotron 3.5 first and selecting it in Settings; downloading a streaming model alone does not activate the feature.
Summaries remain a second layer Muesli itself recommends checking generated notes against the raw transcript before sharing them or using them for decisions, since no transcription or summary system is treated as error free.
Release builds are signed, community builds are not Official releases are signed with the maintainer's Developer ID certificate; external contributions run through a separate unsigned dev build with its own bundle ID.

Quick start

Install Muesli from the DMG in the GitHub Releases, or set it up with brew install --cask muesli.
On first launch, complete the setup wizard and grant Accessibility, Input Monitoring, and recording permissions.
Download a speech recognition model in the Models tab; Parakeet v3 is the recommended default.
Test the dictation hotkey: hold the hotkey, speak, release, and the recognized text appears at the cursor position.
For meetings, optionally connect Google Calendar or the local Calendar app so upcoming events with a meeting link are recognized.

Tips

To avoid false meeting detection from the camera alone, rely on one of the recognized meeting apps; camera use on its own, for example in Photo Booth, does not trigger automatic meeting detection.
Keep only one copy of Muesli in /Applications and delete duplicates, otherwise dictation and hotkey permissions can fail for no obvious reason.
With both local Calendar and Google Calendar connected at the same time, Muesli automatically deduplicates overlapping events, so no manual cleanup is needed.
For summaries without a personal API key, use sign in with an existing ChatGPT Plus or Pro subscription instead of setting up a separate OpenAI key.

Access

Muesli is open source and available on GitHub in the Muesli-HQ/muesli repository, either as a signed DMG from the Releases, through brew install --cask muesli, or as a self built dev build from macOS 14.2 and Xcode 16 onward. The source names no license fee for the app itself; optional summary features through OpenAI or OpenRouter instead run on a separate API key from those providers and are billed outside of Muesli.

Alternatives

fathom.video — freemium

Fathom

Fathom serves the same meeting-notes category as a hosted service with a free entry tier, though it uses a bot participant and cloud processing instead of local recording.

granola.ai — freemium

Granola

Granola follows the same bot-free approach as Muesli, but remains a hosted service with a free basic plan rather than a locally installed app.

otter.ai — freemium

Otter.ai

Otter.ai offers a free allowance of 300 minutes per month, a well-known cloud notetaker as an alternative to Muesli's local processing.

Last reviewed: 27 July 2026 · Pricing, plans and features are a snapshot in time. Check the provider's own page before deciding.

In the workshop this becomes your method.

Whoever sees this process run once wants the agent behind it next. We build that in the workshop From Process to Agent.

View workshops

Prepares your recurring customer sync by consolidating open action items, checking project status, and delivering a compact agenda with prepared answers to expected customer questions.

Learn more

Skills

The Meeting Briefing

Turns meeting type, objective and participant list into a complete briefing document: an agenda with time blocks, anticipated questions and objections with answers, talking points, key facts and a preparation checklist.

Learn more

Skills

The Meeting Framework

Turns meeting notes or a transcript into a structured minutes document with decisions, action items and a ready-to-send follow-up email.

Learn more

Workflows

The Meeting Bridge

A workflow that automatically retrieves the transcript after a meeting ends, builds a structured memo with key topics, decisions, and action items, and sends or drafts a follow-up email to all participants.

Learn more

Assistants

The Meeting Navigator

Prepares meetings in a structured way, delivering a clear agenda with time planning, anticipated questions and objections with prepared answers, and a follow-up checklist for project meetings, strategy sessions, workshops, client meetings and kickoffs.

Learn more

Assistants

The Meeting Minutes Writer

Turns unstructured meeting notes, transcripts or bullet points into clear, scannable minutes with decisions, owners and deadlines.

Learn more

Conversation, not pitch

Understand first, then decide. We take time for an initial conversation, without sales pressure, without obligation.

Schedule a call