Murmur

An open-source AI voice dictation desktop app. A free alternative to Wispr Flow.

Features

• Voice-to-text dictation - Press a hotkey, speak, and your words are transcribed and pasted into any application
• Multiple transcription providers - Choose from Groq, OpenAI, Mistral, or local Whisper models
• AI text processing - Optionally clean up or polish transcriptions using LLMs (OpenAI, Anthropic, Google Gemini, Groq, Mistral, or local Ollama)
• Privacy-first local option - Run entirely offline with local Whisper models
• Cross-platform - Built with Electron for Windows, macOS, and Linux

How It Works

1. Press the global hotkey to start recording
2. Speak your text
3. Release to stop recording
4. Audio is transcribed using your chosen provider
5. Text is optionally processed by an LLM (clean up filler words, polish for clarity)
6. Result is automatically pasted into your active application

Installation

From Source

# Clone the repository
git clone https://github.com/newtro/Murmur.git
cd murmur

# Install dependencies
npm install

# Start in development mode
npm start

Building

# Package the app
npm run package

# Create platform installers
npm run make

Configuration

Recommended: Mistral AI (Free Tier)

If you're new to Murmur, Mistral AI is a great place to start. Their free Experiment plan gives you API access for both transcription and text processing — no credit card required, just a verified phone number.

1. Sign up at console.mistral.ai
2. Activate the Experiment plan (free)
3. Generate an API key
4. In Murmur, select Mistral as both your transcription and LLM provider and paste your key

This lets you use Voxtral for transcription and models like Mistral Small for text processing at no cost. Note that API requests on the free plan may be used by Mistral to improve their models.

Transcription Providers

Provider	Description	API Key Required
Groq	Fast cloud transcription with Whisper	Yes
OpenAI	GPT-4o transcription or Whisper-1	Yes
Mistral	Voxtral cloud transcription	Yes (free tier available)
Local	On-device Whisper (privacy-first)	No

LLM Providers (for text processing)

Provider	Models	API Key Required
Groq	Llama 4 Scout, Llama 3.3 70B	Yes
OpenAI	GPT-4o, GPT-4o Mini	Yes
Anthropic	Claude Haiku 4.5, Sonnet 4.5, Opus 4.5	Yes
Google Gemini	Gemini 2.5 Flash, Gemini 2.5 Pro	Yes
Mistral	Mistral Small, Medium, Large	Yes (free tier available)
Ollama	Any local model	No

Processing Modes

• Raw - No processing, just transcription
• Clean - Remove filler words, add punctuation
• Polish - Full rewrite for clarity and flow

Development

npm start              # Start in development mode with hot-reload
npm run lint           # Run ESLint
npm run typecheck      # Run TypeScript type checking

Architecture

Built with:

• Electron - Cross-platform desktop framework
• React - UI components
• TypeScript - Type safety
• Vite - Fast builds and HMR
• JSON file store - Local settings and history storage
• uiohook-napi - Global hotkey capture

License

MIT

Acknowledgments

Inspired by Wispr Flow - if you want a polished commercial solution, check them out!