explaingit

bshk-app/murmur

Analysis updated 2026-05-18

11SwiftAudience · vibe coderComplexity · 2/5LicenseSetup · moderate

TLDR

A free macOS menu-bar app that turns your spoken words into typed text in any app, running entirely on-device with no cloud or account required.

Mindmap

mindmap
  root((murmur))
    What it does
      Push to talk dictation
      Types in any app
      30 language support
    How it works
      Two on-device models
      Fast draft then refine
      MLX on Apple Silicon
    Privacy
      Fully offline
      No account needed
      Opt-in analytics only
    Setup
      Homebrew install
      Self-updating via Sparkle
      3.6 GB models on first run
Click or tap to explore — scroll the page freely

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

What do people build with it?

USE CASE 1

Dictate code comments, commit messages, or documentation without typing

USE CASE 2

Write messages in Slack, email, or notes apps hands-free while offline

USE CASE 3

Use voice input in any macOS app without giving your audio to a cloud service

USE CASE 4

Switch between two languages mid-sentence for bilingual dictation workflows

What is it built with?

SwiftMLXApple SiliconTuistSparkle

How does it compare?

bshk-app/murmurbootuz/keywordistakageroumado/refrax-browser
Stars111112
LanguageSwiftSwiftSwift
Setup difficultymoderatemoderatehard
Complexity2/53/54/5
Audiencevibe coderdeveloperdeveloper

Figures from each repo's GitHub metadata at analysis time.

How do you get it running?

Difficulty · moderate Time to first run · 30min

Downloads two on-device speech models (~3.6 GB) on first launch, requires macOS 15 and Apple Silicon.

Use freely for any purpose including commercial use as long as you keep the copyright notice.

In plain English

Murmur is a small macOS menu-bar app that lets you dictate text into any app by holding a keyboard shortcut. You hold the key combination, speak naturally, and your words appear wherever your cursor is: in Slack, a code editor, a terminal, a notes app, or anything else. The whole process runs on your Mac with no internet connection required. The app uses two on-device speech recognition models that run through a framework called MLX, which is designed for Apple Silicon chips. When you start talking, a lighter model begins typing almost immediately so you see results right away. A more accurate model then catches up a moment later, quietly correcting names, punctuation, and words that sound alike. You get clean text without waiting for any processing step to finish. Murmur supports around 30 languages and can handle switching between two languages within a single recording. It requires macOS 15 (Sequoia) or later and an Apple Silicon Mac (M1 or newer). On first install, it downloads two models totaling about 3.6 gigabytes, which it stores locally after that. Because everything runs on your device, your voice recordings are never sent anywhere. There are no accounts, no cloud upload, and no stored audio. An optional analytics feature can report anonymous usage errors, but it is turned off by default and never touches audio or transcripts. Builds compiled from source disable analytics entirely. You can install Murmur using Homebrew or download a prebuilt app from the releases page. The app updates itself in the background using a built-in update mechanism called Sparkle. The source code is MIT-licensed and organized into a core library (MurmurKit) that handles audio capture and recognition, with a thin app layer on top for the menu-bar interface and settings.

Copy-paste prompts

Prompt 1
I want to build a macOS menu-bar app in Swift using Tuist that runs a local ML model. Show me how Murmur's MurmurKit structure separates the dictation core from the UI layer.
Prompt 2
How does Murmur's dual-model approach work, where a fast draft model types first and a more accurate model refines the output? Walk me through the VAD and audio pipeline.
Prompt 3
I want to add push-to-talk dictation to my macOS app using MLX and mlx-audio-swift. What permissions are needed and how does the Accessibility permission fallback work?
Prompt 4
Show me how Murmur implements its opt-in PostHog analytics so that source builds have tracking disabled and no audio or transcripts are ever sent.

Frequently asked questions

What is murmur?

A free macOS menu-bar app that turns your spoken words into typed text in any app, running entirely on-device with no cloud or account required.

What language is murmur written in?

Mainly Swift. The stack also includes Swift, MLX, Apple Silicon.

What license does murmur use?

Use freely for any purpose including commercial use as long as you keep the copyright notice.

How hard is murmur to set up?

Setup difficulty is rated moderate, with roughly 30min to a first successful run.

Who is murmur for?

Mainly vibe coder.

Open on GitHub → Explain another repo

This repo across BitVibe Labs

Scan in gitsafehub Deploy in gitdeployhub bshk-app on gitmyhub

Verify against the repo before relying on details.