Crystallize your cognition at the speed of thought
An open-source, system-wide voice typing app for Android. Use Groq Whisper for cloud-powered transcription, or switch to 100% on-device offline mode with AI Agent Mode for intelligent voice commands. No internet required.
Fluence Test Run
The Dictation Bottleneck
Traditional dictation systems are either slow, unpunctuated, or restricted to a specific custom keyboard. Google's streaming STT dictates word-by-word and frequently gets homophones wrong because it lacks contextual sentence-level comprehension. Furthermore, closed-source keyboards pose a massive telemetry and data privacy risk.
The Fluence Approach
Fluence waits for you to finish your sentence/phrase, transcribes it using OpenAI's Whisper Large v3 model on Groq's LPU inference hardware in less than a second, and automatically injects fully-punctuated, context-aware text directly at your cursor.
Accuracy & Performance Comparison
| Feature | Fluence (Groq Whisper v3) | Fluence (SenseVoice Offline) | Google Speech-to-Text |
|---|---|---|---|
| Accuracy | 92% – 97.9% (Human level) | 85% – 93% (On-device) | 79% – 88% |
| Punctuation | Automatic & Intelligent | Automatic & Intelligent | Rigid / Word-by-word |
| Privacy | Open Source (Zero Telemetry) | 100% On-Device (Zero Egress) | Closed Source (Data collected) |
| System-wide Overlay | Yes (Floating Bubble) | Yes (Floating Bubble) | No (Restricted to keyboard) |
| Accent & Jargon Handling | Excellent (680k hr dataset) | Good (Industrial dataset) | Average (Often stumbles) |
| Internet Required | Cloud only | No. Fully Offline. | Cloud only |
Setup Wizard
Crystallize your cognition at the speed of thought
⚡ Quick Install & Setup
- Download APK: Get the latest version from GitHub Releases and install it on your Android device.
- Configure API: Open the Fluence app and paste your Groq API Key (stored securely on-device with hardware encryption). Not needed for offline mode.
- Enable Accessibility Service: Turn on the Fluence Accessibility Service inside your system Accessibility settings so Fluence can detect text field focus events.
- Grant Overlay: Enable the Draw Over Other Apps permission to show the floating bubble near your cursor.
- Offline Download: Requires a one-time download of the SenseVoice-Small model (~230 MB) within the app settings to enable 100% offline transcription.
- Single Tap: Starts or stops standard voice transcription.
- Hold / Long Press: Activates AI Agent Mode to process natural language voice commands.
API Configuration
Requires a Groq API Key to perform local-to-cloud transcription securely. Skip this step if you only use offline mode.
Microphone Access
Grant audio recording access to enable speech-to-text dictation.
Enable Accessibility Service
Activate the Fluence Accessibility Service under your system Accessibility settings so it can detect focused text fields.
Engineered for Perfection
Voice Dictation
Powered by Groq-accelerated Whisper Large v3. Transcribes in under a second with up to 97.9% word accuracy, handling complex jargon and heavy accents with ease.
AI Agent Mode
Hold to activate. Powered by Llama 3.3 70B, Agent Mode accepts speech instructions right at your cursor: edit, rewrite, translate, or generate text in any app.
Ironclad Privacy
Zero telemetry, zero keyloggers, zero analytics. Audio is ephemeral and deleted instantly after transcription. Your API Key is hardware-encrypted via Android Keystore.
Local Edge Model
100% offline voice typing powered by Alibaba's SenseVoice-Small ONNX model (~230 MB, downloaded on demand). All inference runs on-device via sherpa-onnx. No internet, no data egress, ever.
Future Roadmap
Desktop Release
Bringing system-wide voice dictation and AI agent assistance to desktop. Plans include a cross-platform app for Windows, macOS, and Linux with global keyboard triggers, clipboard injection, and local LLM support.
Provider Flexibility
Support for any OpenAI-compatible API endpoint, so you are never locked into a single provider. Bring your own key from Mistral, OpenRouter, Cerebras, or any other compatible service. Additional offline models will also be added based on user feedback.
Frequently Asked Questions
Is Fluence free to use?
Yes, Fluence is 100% free and open-source under the Apache 2.0 license. Since transcription is processed using API keys, you only pay for your own usage. Groq offers a generous free tier for developers, and even on paid tiers, Whisper transcription costs a fraction of a cent per hour of usage. Offline mode using SenseVoice has no API cost whatsoever.
Does Fluence track my keystrokes or usage?
Absolutely not. Fluence does not record, store, or transmit keystrokes, typing speed, session length, usage frequency, or any behavioral data whatsoever. There are zero analytics libraries, zero crash reporters, and zero background telemetry beacons in the codebase. The app contains no tracking code of any kind. Every line of this is verifiable since the project is fully open source.
How is my data privacy protected?
Fluence is built on ironclad privacy principles. There is zero telemetry and no server logging of any kind. Audio is recorded, sent securely to the Whisper API, transcribed, and instantly discarded from both the API and your device. Your Groq API key is stored inside the Android Keystore using hardware-backed encryption, never in plain text. In offline mode, no audio or data ever leaves your device at all.
Can it run completely offline?
Yes! Fluence fully supports 100% offline, on-device voice dictation. Toggle "Offline Transcription" in settings and the app downloads Alibaba's optimized SenseVoice-Small ONNX model (~230 MB, one-time download). Once downloaded, all transcription runs completely locally on your device with zero internet connectivity and zero data egress. Offline mode also works with AI Agent Mode for local text editing commands.
Does it support languages other than English?
Yes. Whisper Large v3 supports 99+ languages with high accuracy and automatic language detection. Just speak in your native language and it transcribes correctly. SenseVoice offline mode also supports multiple languages including Chinese, Japanese, Korean, and more, with automatic detection.