PromptFlux

Local, offline voice-to-text for Windows.

Press a shortcut, speak, and get text in your clipboard or pasted into your active app. No cloud calls for transcription, no accounts, no telemetry.

Why PromptFlux

• Local-first speech recognition using faster-whisper
• Low-latency pre-buffered recording to avoid clipped first words
• Multiple trigger modes:
  • Hold-to-talk
  • Press-and-release to speak (auto-stop on silence)
  • Wake word
• Optional mobile relay over local network (HTTPS + token)
• Windows installer includes packaged STT service (no Python needed on target machine)

Keywords

offline speech to text, local whisper app, voice typing windows, dictation desktop app, hold to talk transcription, wake word transcription, private speech recognition

Requirements

• Windows 10/11
• Microphone
• Approx. 200MB disk (app + model)

Installation

1. Download the latest installer from GitHub Releases.
2. Install and launch PromptFlux.
3. On first run, the Whisper model is downloaded once.

Installer output filename:

• PromptFlux Setup 0.1.0.exe

Quick Start

1. Open PromptFlux.
2. Open Settings.
3. Choose trigger mode:
   • Hold To Talk (Hotkey)
   • Press And Release (Hotkey)
   • Wake Word
4. Save with Confirm.
5. Speak and receive text in clipboard or auto-paste mode.

Trigger Modes

Hold To Talk

• Press and hold hotkey to record
• Release to transcribe

Press And Release

• Tap hotkey once to start recording
• Speak naturally
• Recording stops automatically on silence (or max duration)

Wake Word

• Say configured wake phrase
• Recording starts automatically
• Stops on silence (or max duration)

System Audio Capture

PromptFlux supports system audio capture through WASAPI. Depending on your PortAudio/sounddevice backend, direct loopback on output devices may not be available.

If needed, choose a [Input Capture] style device (for example Stereo Mix, Voicemeeter Out, virtual cable) in settings.

Configuration

File location:

• %APPDATA%/promptflux/config.json

Common settings:

• hotkey
• forceEndHotkey
• triggerMode
• wakeWord
• wakeSilenceMs
• wakeRecordMs
• outputMode
• captureSource
• microphoneDevice
• systemAudioDevice

Architecture

Electron app (UI, hotkeys, clipboard, watchdog) communicates with a local Python STT service over localhost WebSocket.

See:

• docs/ARCHITECTURE.md
• docs/DECISIONS.md

Development

Prerequisites

• Node.js 20+
• Python 3.11+

Setup

cd electron-app
npm install

cd ../stt-service
pip install -r requirements.txt

Run (dev)

# Terminal 1
cd stt-service
python server.py

# Terminal 2
cd electron-app
npm start

Build Windows Installer

cd electron-app
npm run dist:win

Packaging

• Electron app: electron-builder
• STT service: PyInstaller via stt-service/build_windows.ps1
• Final artifacts: electron-app/dist/

Security and Privacy

• Transcription runs locally
• No cloud API for speech recognition
• WebSocket bound to 127.0.0.1

See SECURITY.md for full policy.

Contributing

Please read:

• CONTRIBUTING.md
• CODE_OF_CONDUCT.md

License

This repository is licensed under GNU Affero General Public License v3.0.

• Full license text: LICENSE
• Commercial licensing notes: LICENSE-COMMERCIAL.md

Roadmap

See ROADMAP.md.

Support

• Bug reports: GitHub Issues
• Security reports: see SECURITY.md