OpenWhispr/openwhispr | Repo of the Day

Voice-to-text dictation app with local (Nvidia Parakeet/Whisper) and cloud models (BYOK). Privacy-first and available cross-platform

Summary

OpenWhispr is an open-source, cross-platform desktop app that turns voice into text, notes, and commands, with the choice of running local speech models (Whisper, NVIDIA Parakeet) or bringing your own cloud API keys. It is built on Electron 41 and React 19, uses whisper.cpp and sherpa-onnx for on-device transcription, and exposes a public API plus an MCP server for programmatic access.

What it is useful for

Dictating into any focused text field via a global hotkey, with automatic pasting, plus a separate hotkey for translating speech between languages.
Transcribing meetings on Zoom, Teams, and FaceTime with on-device speaker diarization and voice fingerprinting, plus Google, Microsoft, or Apple Calendar integration.
Talking to AI assistants (GPT-5, Claude, Gemini, Groq, Tinfoil, OpenRouter, or local models) by voice, with optional screenshot context, and having highlighted text edited in place.
Managing notes with folders, semantic search, cloud sync, team spaces, and link-based sharing.
Importing existing audio or video for transcription, including batch uploads and YouTube URLs, with optional speaker detection.
Driving notes and transcriptions from code via the public API, or from another AI assistant via the MCP server.
Enterprise scenarios that need policy enforcement, SSO/SCIM, or centrally managed Amazon Bedrock or Azure OpenAI access without distributing keys to users.

How engineers can use it

Run it locally with the documented quick start: git clone https://github.com/OpenWhispr/openwhispr.git, cd openwhispr, npm install, npm run dev. The README states Node.js 24+ is required. The desktop app stays at the repository root; the Expo mobile app lives in openwhispr-mobile/ with its own dependencies and README.
Choose a backend: use whisper.cpp and sherpa-onnx for offline transcription, with GPU acceleration on Metal, CUDA, and Vulkan (including AMD/Intel), or plug cloud provider keys in. A custom-asr-shim example under examples/custom-asr-shim/ shows how to wire in non-OpenAI-compatible ASR APIs for self-hosted transcription.
Documented limitation: on Intel Macs, live speaker identification and voice fingerprinting are unavailable because ONNX Runtime stopped shipping macOS x86_64 binaries in 1.24. Meeting recording and basic transcription still work, and notes search falls back to keyword matching instead of semantic search. Platform-specific setup, build details, and troubleshooting live on the docs site linked in the README.