Skip to content

Repository files navigation

CrunchyMurmur logo

CrunchyMurmur

Private-by-default voice dictation, meeting transcription, and Markdown notes.

Press a shortcut, speak, and keep working. CrunchyMurmur runs on Windows, macOS, and Linux, with fast local Parakeet, broad-language Whisper, or optional cloud transcription.

Latest release Downloads CI CodeQL CodeRabbit Pull Request Reviews Platforms License: MIT

Website · Download · Documentation · Report a bug · Security

CrunchyMurmur dashboard with live dictation active


What it does

Capability What you get
Dictation anywhere Hold the global shortcut, speak, and release to transcribe into the focused app.
Meeting capture Record a microphone and, where supported, system audio as separate speaker-labelled tracks.
A real Markdown workspace Edit notes, meeting notes, and reusable AI templates with write, split, and preview modes.
Local-first privacy Use whisper.cpp fully offline. There is no account, advertising SDK, or project-operated telemetry.
Useful progress, locally Track dictated words, weighted words per minute, and consecutive active days on your device.
Verified releases and updates Signed packages, checksums, SBOMs, provenance attestations, and stable in-app updates from GitHub Releases.

The interface is available in 12 languages — English, Italiano, Español, Português, Français, Deutsch, Dansk, Norsk, Svenska, 中文, 한국어, and 日本語. It follows your system language by default, and you can switch instantly in General settings.

A calm, native-feeling workspace

Appearance and device settings Local and cloud transcription
CrunchyMurmur General settings in light mode CrunchyMurmur transcription engine settings

Screenshots use a clean automated test profile with no personal data or credentials.

Notes that stay yours

Markdown notes workspace Focused note editing
CrunchyMurmur Notes workspace CrunchyMurmur Markdown note editor

Notes are ordinary Markdown files in your Documents folder, so they remain portable, searchable, and usable with other editors.

Install

Download the latest package from GitHub Releases, or use the verified terminal installer.

Windows — PowerShell

irm https://raw.githubusercontent.com/a-streetcoder/CrunchyMurmur/main/install.ps1 | iex

macOS or Linux

curl -fsSL https://raw.githubusercontent.com/a-streetcoder/CrunchyMurmur/main/install.sh | sh

The scripts detect your operating system and architecture, download the latest release, and verify it against the published SHA-256 digest before installation. See Getting started for manual packages and first-run setup.

Windows trust notice: Windows installers are Authenticode-signed with the temporary self-signed publisher certificate CN=CrunchyMurmur Temporary Self-Signed Publisher, valid through 10 July 2027. SmartScreen may warn because the certificate is not issued by a publicly trusted certificate authority. Verify SHA256SUMS and the GitHub provenance attestation before installing. macOS packages are Developer ID signed and notarized.

Choose where transcription happens

  • Parakeet V3 (recommended): fast, accurate CPU transcription for 25 European languages through the bundled Rust transcribe-rs engine. The model stays loaded between dictations and meeting chunks.
  • Local Whisper: the broad-language fallback for 99+ languages, manual language selection, translation, and custom GGML models. CrunchyMurmur bundles whisper.cpp; no separate engine installation is required.
  • Groq Whisper API: faster on many laptops; audio is sent to the Groq account you configure.

Near-silent recordings are rejected before inference instead of being saved as model hallucinations.

  • Optional AI formatting and notes: use Anthropic, OpenAI, Groq, Claude Code, or Codex only when you configure and invoke them.

Cloud features are optional. Provider failures fall back safely without discarding the original transcript.

Designed for trust

  • No CrunchyMurmur account, analytics SDK, ads, or project-operated telemetry.
  • API keys use operating-system-backed safe storage when available.
  • Local data can be exported or permanently deleted from the app.
  • Audio retention never removes meeting transcripts or notes.
  • Release artifacts include SHA-256 checksums, an SPDX SBOM, and GitHub build-provenance attestations.
  • Privileged Electron operations stay in the main process with constrained renderer APIs.

Read the complete privacy notice, security policy, and architecture overview.

Platform support

Windows macOS Linux
Packages NSIS x64 / ARM64 Universal DMG / ZIP AppImage and Debian x64 / ARM64
Default shortcut Hold Ctrl + Win Hold Fn Configurable toggle
Meeting system audio Supported macOS 13+ Microphone only
Automatic paste Native Accessibility permission wtype or xdotool
Updates Automatic Automatic Automatic for AppImage; reinstall for Debian

See the full capability matrix and Linux requirements.

Updating

Packaged Windows, macOS, and Linux AppImage builds check stable GitHub Releases and download valid upgrades in the background. The app asks before restarting to install. Prereleases and downgrades are never applied.

You can also run the same terminal install command again at any time. Debian packages use this manual path. Read Updating CrunchyMurmur for exact behavior and verification options.

Build from source

Requires Node.js 22.12 or newer, npm, the stable Rust toolchain, and a native C/C++ linker (Visual Studio Build Tools on Windows, Xcode Command Line Tools on macOS, or GCC on Linux). Git is optional: the source bootstrap resolves main to an exact commit, downloads its GitHub archive, installs locked dependencies, validates the project, builds the local transcription helper, and launches it.

Windows — PowerShell

& ([scriptblock]::Create((irm https://raw.githubusercontent.com/a-streetcoder/CrunchyMurmur/main/scripts/source/run-from-source.ps1)))

macOS or Linux

curl -fsSL https://raw.githubusercontent.com/a-streetcoder/CrunchyMurmur/main/scripts/source/run-from-source.sh | sh

The source is kept in a user-owned application-data directory, so running the command again safely rebuilds from the latest protected main. The previous working source tree is replaced only after the new commit passes dependency installation and validation. Read the scripts before piping them to a shell; options and manual commands are documented in Building from source.

Manual development setup remains available:

git clone https://github.com/a-streetcoder/CrunchyMurmur.git
cd CrunchyMurmur
npm ci
npm run check
npm start

Native release packages must be built on their target platform. See Contributing and Releasing.

On-device transcription SDK

The local Parakeet engine is also available as a desktop SDK release candidate for Node/Electron and Tauri. It exposes a small shared lifecycle: prepare, transcribe, diagnostics, and dispose. Microphone permission and capture stay with the host application, while model validation, warm reuse, inference, stable errors, and privacy-safe diagnostics stay in the shared Rust engine.

The Electron chat demo is a complete integration example with push-to-talk recording and a secure renderer-to-main boundary. Read the SDK guide for installation, platform scope, trust requirements, publication status, and the Swift/Kotlin/React Native roadmap.

Documentation

Contributing

Bug reports, focused pull requests, documentation improvements, and platform testing are welcome. Please read Contributing and the Code of Conduct first. Never post API keys, private calendar URLs, recordings, transcripts, or notes in a public issue.

CrunchyMurmur is released under the MIT License. Use is also subject to the distributed Terms of Use.

The Markdown editing experience is powered by MarkText's MIT-licensed Muya editor. See third-party notices for attribution details.

About

Private-by-default cross-platform voice dictation, meeting transcription, and Markdown notes.

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages