Muesli is a free, open-source dictation app for Apple Silicon Macs running macOS 14.2 or later. Download a local speech model, hold your hotkey, speak, and release to paste into the active text field.
Hold Right Command, speak naturally, and Muesli pastes clean text into the note you were already writing.
The dictation model runs on Apple Silicon. No hosted speech-to-text API is needed for quick thoughts, issue drafts, messages, or notes.
Hold Right CmdMuesli is a local-first macOS dictation app that turns speech into text on your Mac and pastes it into the app you are already using.
For normal dictation, speech recognition runs on-device instead of sending audio to a hosted speech-to-text API.
Use Muesli for notes, email, Slack, docs, issue trackers, prompts, and other Mac text fields without moving your writing into a separate editor.
Dictation is most useful when it disappears into normal work. Muesli does not ask you to move your writing into a recorder app. It listens from the menu bar and pastes into Messages, Slack, Linear, Notion, Google Docs, email, terminals, issue trackers, or wherever the cursor is waiting.
Use a modifier key such as Right Command when you want to dictate. Speak in the app you were already using.
Muesli runs speech recognition locally through CoreML, Metal, and the Apple Neural Engine. Parakeet is built for fast everyday dictation.
When you release the hotkey, Muesli pastes the cleaned text into the active text field instead of making you copy from a separate transcript window.
Some models require newer macOS versions. Secure fields and apps that restrict Accessibility may need manual pasting. Review text before sending it.
For setup help, read the offline dictation guide. Compare the Superwhisper workflow, or use meeting transcription for a call.
Local speech recognition processes audio on your Mac. Hosted dictation is optional and sends audio to OpenAI or through OpenRouter. Hosted cleanup sends text; hosted summaries send meeting content. Optional iCloud sync transfers text and sync metadata, not audio. A local speech model alone does not make every enabled feature offline.
After the model is installed, day-to-day dictation does not need a hosted speech-to-text API. Your voice is processed on your Mac, which removes the usual upload, queue, retention, and vendor-account surface from quick dictation.
Cloud transcription can be the right tradeoff for some teams, but it adds more places where audio, transcripts, credentials, logs, and third-party access policies have to be trusted. Muesli keeps the dictation path narrower.
Muesli is public on GitHub, so the app, model routing, permissions, paste behavior, and local storage choices can be inspected instead of hidden behind a hosted black box.
The recommended path is Parakeet on the Apple Neural Engine for quick dictation. Muesli also supports Whisper and Qwen3 ASR options for people who care more about a particular language, model family, or accuracy tradeoff than raw speed.
Time from releasing the hotkey to pasted text depends on your Mac, model, recording length, model warmup and cleanup settings. There is no controlled latency benchmark for these configurations on this page. Try the same short draft with each downloaded model before choosing.
Short answers for the parts people usually want clarified before trusting dictation software.
Local dictation processes audio on your Mac. Selecting hosted dictation sends audio to OpenAI or through OpenRouter. Hosted cleanup sends text, and hosted meeting summaries send meeting content to the selected provider. Check each model setting before use.
Yes. Muesli is designed as a local-first Mac dictation app for people who do not want quick speech-to-text to depend on a hosted transcription service.
After downloading a local speech model, dictation works offline. Select local cleanup or turn cleanup off. Hosted dictation, hosted cleanup and hosted summaries require a connection; downloads, updates and optional sync also use the internet.
Muesli pastes dictated text into the active app using macOS accessibility and clipboard behavior, so it is built for notes, email, chat, documents, issue trackers, browser text fields, and other places you already type.
Muesli is open-source, model-flexible, and built around a hold-to-talk workflow with local ASR models such as Parakeet, Whisper, and Qwen3 ASR. It also connects dictation with meeting transcription and local-first notes workflows.
Dictation needs microphone access, input monitoring for the hotkey, and accessibility permission to paste the result. The app guides those permissions during onboarding.
Open-source, Mac-native, and built for people who would rather keep their voice close.
Download Muesli