Talk naturally.
Paste perfectly.

On-device AI dictation for macOS. Record your voice, get polished text — grammar-fixed, filler-free, pasted into any app in under 2 seconds.

Four steps.
Under two seconds.

A streaming pipeline that starts transcribing while you're still talking.

Record

Choose between push-to-talk or toggle. Talk naturally. Release. Done.

Transcribe

Parakeet v3 on Apple Neural Engine at 110x speed. Streaming while you speak.

Polish

LLM removes filler words, fixes grammar. OpenAI, Gemini, Ollama, or Apple Intelligence.

Paste

Polished text auto-pastes into any app. Previous clipboard preserved.

Streaming pipeline
in action.

What you said Streaming ASR
Pipeline
Record
5.0s
Stream ASR
real-time
parallel
Finalize
~80ms
Pre-warm
TLS
SSE Polish
~1.5s
Paste
<5ms
What got pasted SSE Streaming

Built different.

0x
Real-time transcription speed
Parakeet v3 on Neural Engine
0s
Silence to paste
Overlapped streaming pipeline
0 bytes
Audio uploaded
On-device transcription

Everything you need.
Nothing you don't.

Built for people who think faster than they type.

On-device transcription

Parakeet v3 and WhisperKit run entirely on your Mac. No internet required. No audio uploaded. Ever.

Smart polish

LLM-powered cleanup removes filler words, fixes grammar, and formats your text. Editable prompts with presets for Clean Up, Formal, and Casual styles.

Voice activity detection

Silero VAD with smoothed probability tracking, 512ms prebuffer for word onset capture, and configurable silence timeout.

Works in any app

Auto-pastes into whatever app you were using — Slack, VS Code, Gmail, Notes, Terminal. Previous clipboard contents preserved.

Push-to-talk & toggle

Hold a key to record, or press once to start and again to stop. Customizable hotkeys. Cancel anytime with Escape.

Four LLM providers

OpenAI, Google Gemini, Ollama (local), and Apple Intelligence. API keys stored in macOS Keychain. Switch without losing credentials.

Custom word correction

Teach it your jargon. Add names, acronyms, and technical terms. Phonetic and edit-distance matching handles misheard words.

Native macOS app

Built in Swift 6 with SwiftUI. Lives in your menu bar. No Electron. No WebView. Feels like it belongs on your Mac because it does.

Streaming pipeline

Transcription starts while you're still talking. ASR, polish, and paste overlap in parallel — no waiting for each step to finish.

Your voice never
leaves your Mac.

On-device transcription means your audio stays on your hardware. No cloud ASR. No recordings stored on remote servers. No telemetry.

On-device ASR

Both speech engines — Parakeet v3 and WhisperKit — run locally on Apple Silicon. Audio is processed in-memory and never written to disk.

Keychain-secured credentials

API keys for LLM providers are stored in macOS Keychain, not config files. The same security that protects your passwords protects your keys.

Optional LLM polish

The only network call is the optional LLM polish step — and you control the provider. Use Ollama for fully offline operation.

Open source

Every line of code is on GitHub. Audit the source. Verify the claims. No trust required.

EnviousWispr — Privacy
🛡
All audio stays on-device
🎤 Audio capture Local
🧠 Speech-to-text Local
LLM polish Optional
📤 Audio upload Never

Open source.
Open roadmap.

EnviousWispr is built in the open. Read the code, file issues, suggest features, or contribute directly.

Terminal — EnviousWispr
$ git clone https://github.com/saurabhav88/EnviousWispr.git
Cloning into 'EnviousWispr'...
$ cd EnviousWispr && swift build
Building for debugging...
Build complete! (62 sources compiled)
View on GitHub

Contributions welcome. Built with Swift 6, SwiftUI, and Apple's Neural Engine APIs.

Start dictating
in 30 seconds.

1

Download

Grab the latest DMG from GitHub Releases. Drag to Applications.

2

Grant permissions

Microphone access for recording. Accessibility for auto-paste. A guided setup walks you through both.

3

Press your hotkey

Control+Space to toggle. Option+Space to push-to-talk. Talk naturally. Release. Done.