Getting Started with Fluence

Crystallize your cognition at the speed of thought. Fluence is a system-wide voice dictation tool and smart AI editor built for total data privacy.

💡 Core Philosophy

Unlike standard speech-to-text engines that type word-by-word and frequently miss context, Fluence captures your full spoken sentence, refines punctuation dynamically, and injects clean text at your cursor position instantly. It works in any text area, text editor, browser, or coding environment.

Platform Support

Fluence supports two distinct runtime environments designed to integrate natively with your daily workflow:

Windows Application

Powered by Tauri. Sits lightweight in your system tray and triggers via global hotkeys to write into any desktop software.

Explore Windows Features

Android Application

Powered by Jetpack Compose. Utilizes Android Accessibility Services to float a bubble near your active text cursor on mobile.

Explore Android Features

Windows Desktop Guide

Configure and run the Tauri desktop integration for system-wide dictation on Windows devices.

Installation Steps

  1. Go to the Latest Windows Release page.
  2. Download the setup file: select .exe for a standard user setup or .msi for enterprise deployment.
  3. Run the installer and follow the instructions. The application will start and place an icon in your system tray.

Tray Icon and Settings

Right-click the Fluence tray icon (microphone shape) to access settings:

  • Preferences: Opens the AMOLED dashboard where you can define shortcut keys, autostart behavior, and cloud keys.
  • Local Dictionary: Add jargon, personal acronyms, and names to prevent spelling mistakes.
  • View Logs: Review local diagnostic console outputs.
  • Exit: Completely terminate the application.

Hotkey Configuration

The default global hotkey is Ctrl + Shift + Space. When pressed inside any text field, the application will display a compact overlay pill at the bottom of the screen showing a recording wave. Release or press the shortcut again to finish dictation and insert text.

Android Mobile Guide

Setup and configure the mobile app. Learn how to activate Accessibility overlays for system-wide dictation.

Sideloading the APK

  1. Navigate to the Android Releases page on your mobile device.
  2. Download the app-release.apk asset.
  3. Open the file. If prompted, toggle "Allow installation from unknown sources" in Android settings to finalize.

Required System Permissions

⚠️ Crucial Permissions

To function as a universal text injector, you must enable these two system permissions inside the onboarding screens:

  • Accessibility Service: Allows the application to scan focused text boxes so it can locate the position of your cursor.
  • Display Over Other Apps (Overlay): Permits the glassmorphic bubble to float above active third-party apps.

Bubble Interactions

A floating microphone bubble sits next to your active cursor. Tap to dictate, or hold the bubble to speak and release to finish. A bottom voice bar gives real-time visual feedback.

Offline Mode Configuration

Run 100% on-device speech-to-text without sending data online. This is perfect for high-privacy environments or users with poor connectivity.

Alibaba SenseVoice-Small Integration

Fluence integrates the Alibaba SenseVoice-Small model. This quantized model is extremely fast and light, with a size of approximately 230 MB.

Configuration for Android

  1. Ensure you have a stable connection for the initial setup.
  2. Open Fluence Settings and toggle Offline Mode to active.
  3. The application will automatically download the required weights. Do not close the app during the download.
  4. Once the verification check succeeds, you can turn off internet access and dictate locally.

Configuration for Windows

  1. Open the preferences panel from the system tray.
  2. Navigate to the AI Providers tab.
  3. Select Local SenseVoice as your active transcriber.
  4. Click download to fetch the ONNX weights. Once complete, audio files will be processed locally via the sherpa-onnx runtime.

Privacy Policy (Zero Telemetry)

Your data belongs to you. Zero egress, zero telemetry, zero compromises.

🔒 Plain English Commitments

  • No Tracking Scripts: We do not integrate Google Analytics, Firebase, Sentry, or any telemetry framework. We collect no analytics whatsoever.
  • No Audio Egress in Offline Mode: In offline mode, the microphone captures audio data directly into local volatile RAM and deletes it immediately after the local ONNX inference runs. No network requests are made.
  • Direct Cloud Delivery: In cloud mode, audio data goes directly from your local client to the secure Groq API gateway via HTTPS. No intermediate backend acts as a wrapper.
  • Credential Storage: Your cloud model API keys are stored locally using native system keychains: Android Keystore (using hardware-backed EncryptedSharedPreferences) and Windows Credential Manager.
  • Auditable Codebase: Every line of code is open source and auditable. Feel free to inspect, compile, or run it yourself.

Troubleshooting & FAQs

Common issues and how to resolve them.

1. The Android bubble is not appearing in certain apps

Make sure the Accessibility Service is toggled on in your system settings. Certain high-security text boxes, like banking password fields, block accessibility helpers by design to prevent overlay attacks.

2. I get a transcription error in Cloud Mode

Check your internet connection first. If online, open settings and confirm your Groq API key is valid and has no trailing whitespaces. If you trigger multiple short dictations in rapid succession, you might temporarily hit Groq's rate limits.

3. How much storage does Offline Mode require?

The SenseVoice model and Silero VAD weights require about 230 MB of local storage space. Ensure your phone or desktop C: drive has adequate free capacity before toggling the download.

4. Why does the Windows overlay pill not have close buttons?

Clicking overlay buttons on Windows forces desktop systems to transfer active window focus away from your editor. To keep typing fluidly in your main app, the Windows pill displays status only and does not capture input focus.