OpenWhispr/openwhispr

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

OpenWhispr – What it is

OpenWhispr is an open‑source desktop application that turns spoken words into text and lets you interact with AI assistants right from your computer. It works on macOS, Windows and Linux and gives you a choice between offline transcription (using local Whisper‑based engines) and cloud services for faster results. All core data stays on your device unless you opt‑in to cloud features, so privacy is a first‑class concern.

Core capabilities

Capability	How it works	What you get
Voice dictation	Press a global hot‑key, speak, and the transcribed text is automatically pasted wherever the cursor is.	Hands‑free typing in any app.
Translation while dictating	Same hot‑key, but the spoken language is translated on‑the‑fly.	Multilingual dictation without leaving the keyboard.
AI‑powered assistant	Sends your spoken prompt to a chosen LLM (GPT‑5, Claude, Gemini, Groq, Tinfoil, OpenRouter, or a locally‑run model).	Conversational answers, code help, summarisation, etc., spoken or inserted as text.
Meeting transcription	Detects Zoom, Teams, FaceTime calls, runs live speaker diarisation and voice‑fingerprinting.	Real‑time, searchable transcripts with speaker labels.
Notes & semantic search	Stores transcriptions/notes in a local SQLite DB, optionally syncs to the cloud, and indexes with embeddings for semantic search.	Quick retrieval of past content, even by meaning rather than keywords.
Team spaces	Cloud‑based sharing with role‑based permissions, SSO/SCIM integration.	Collaborative note‑taking for groups or enterprises.
Audio import	Drag‑and‑drop files, batch upload, or paste YouTube/audio URLs.	Transcribe existing recordings with optional speaker detection.
Enterprise controls	Centralised policy, SSO, and managed access to Bedrock or Azure OpenAI keys.	Secure rollout inside organisations.
Public API & MCP server	REST endpoints for notes, transcriptions, and a “MCP” integration point for custom AI assistants.	Programmatic access and extensibility.

Technology stack (as described in the repo)

Frontend: React 19 + TypeScript + Tailwind CSS v4, UI built with shadcn/ui.
Desktop shell: Electron 41.
Local speech‑to‑text: whisper.cpp (CPU/GPU via Metal, CUDA, Vulkan) and sherpa‑onnx for NVIDIA Parakeet.
Speaker diarisation / fingerprinting: ONNX‑runtime models.
LLM inference: llama.cpp for running local LLMs; cloud LLMs accessed via their APIs.
Data storage: better-sqlite3 (embedded SQLite).
Cloud backend: Neon serverless Postgres powers the optional OpenWhispr Cloud.

Getting started (quick‑start from the README)

# Clone the repo
git clone https://github.com/OpenWhispr/openwhispr.git
cd openwhispr
# Install dependencies (Node 24+ required)
npm install
# Run the development build
npm run dev

For end‑users, pre‑built binaries are provided:

macOS (.dmg) – Apple Silicon and Intel (note: some speaker‑ID features need ONNX which is unavailable on Intel macOS).
Windows (.exe)
Linux (.AppImage, .deb, .rpm, .tar.gz)

Who might benefit?

Knowledge workers who want fast, private dictation without switching apps.
Multilingual professionals needing on‑the‑fly translation while typing.
Teams that record virtual meetings and need searchable transcripts with speaker labels.
Developers / power users who want to hook the public API or the MCP server into their own workflows.
Enterprises looking for a self‑hosted, policy‑compliant voice‑to‑text solution.

License & community

Licensed under the MIT license – free for personal and commercial use.
Contributions are welcomed via pull requests; a contributing guide is linked in the docs.
The project is actively maintained (see release badge, star history, and sponsor Neon for cloud services).

Bottom line

OpenWhispr bundles modern speech‑recognition (Whisper, Parakeet), on‑device speaker diarisation, and LLM‑backed assistants into a single cross‑platform desktop app. Its design lets you stay offline for privacy‑critical work or switch to cloud APIs for speed, and it offers extensibility through a public API and an MCP integration point.