Murmur

What to know before you start

Murmur is an open-source macOS menu-bar application for push-to-talk voice dictation. The basic interaction is direct: hold a configured key, speak, and release it to send the recording for transcription. Murmur then pastes the resulting text at the active cursor. Its repository documents use across terminals, code editors, browsers, messaging applications, Notes, and other text fields. Terminal insertion is handled as a single bracketed paste, which is intended to prevent a multi-line dictation from being executed one line at a time. Users can also configure tap-to-toggle recording and press Escape to cancel an active dictation.

The application sends recorded audio directly from the Mac to OpenAI and requires the user to supply an OpenAI API key. According to the project documentation, Murmur has no user account, subscription, or intermediary Murmur server. OpenAI API usage is billed separately from a ChatGPT subscription, so users need an API account with available credit or a payment method. The repository estimates transcription at approximately $0.27 per hour of speech, while the creator reported a typical personal daily spend of about $0.15 in the Show HN post. These figures are author-provided estimates rather than independently verified usage benchmarks, and actual costs can vary with usage and the OpenAI services selected.

Murmur includes several controls for people who dictate frequently. A shortcut can use a key combination or, with the necessary macOS permission, a single key such as Fn or Right Option. Recording begins when the shortcut is pressed, and users can begin another dictation while an earlier recording is still being transcribed. The project says completed results are inserted in spoken order. Failed transcription attempts retain their recordings and can be retried. Automatic language detection supports multilingual speech and switching languages within a sentence, while a custom vocabulary can help with names, jargon, and product-specific terms.

Modes extend Murmur beyond literal speech-to-text. Each Mode contains instructions for an OpenAI model to transform a transcript before insertion. Documented examples include translating speech into English, removing filler words, producing bullet points, or turning spoken content into a commit message. Modes can have their own shortcuts, allowing different transformations to be selected without repeatedly entering the same prompt. This processing remains cloud-based. Murmur does not provide local speech recognition, and its repository explicitly directs people who require entirely on-device transcription toward other projects.

Installation is more involved than installing a conventional signed Mac application. Murmur does not currently offer a prebuilt download, so users need macOS 14 Sonoma or later, full Xcode 16 or later, Git, and the ability to run the supplied build script. The documented process clones the repository, runs Scripts/build.sh --install, copies the resulting application into /Applications, and opens it. Initial setup requires Microphone permission, Accessibility permission so Murmur can paste into other applications, and an OpenAI API key stored through the application’s settings. The repository states that Accessibility access is used for insertion and that Murmur does not read the screen or keystrokes.

There are operational limitations to weigh before adopting it. Default ad-hoc signing may cause macOS to treat an updated build as a new application, requiring Microphone and Accessibility permissions to be granted again. Signing with an Apple Development certificate can avoid that repeated permission process, but it introduces another setup step. Single-key shortcuts do not work while a password field is focused, and macOS 15 restricts some shortcuts that use only Option or Option-Shift modifiers. Editorially, Murmur is best evaluated as a source-built utility for technically comfortable Mac users who value a transparent bring-your-own-key architecture. It is less suitable for people seeking a packaged download, offline transcription, mobile or Windows support, or a product that abstracts away API billing and macOS security permissions.