Word Tone is a high-performance, edge-integrated AI transcription and rewriting engine delivered as a custom Android Input Method. It leverages large language models (LLMs) to provide real-time syntactic refinement, tone adjustment, and grammatical sanitization across the entire Android ecosystem.
Word Tone is designed as a distributed system, separating high-latency generative AI computation from the low-latency interactive keyboard interface.
Word Tone in action: QWERTY Layout, AI Tone Selection, and Real-time Suggestions.
- IMF Integration: Inherits from
InputMethodServiceto interact with the Android Input Method Framework, ensuring global accessibility across all input fields. - Dynamic UI Inflation: Utilizes native binary XML inflation to minimize view creation latency and memory overhead during layout transitions.
- Asynchronous Processing: All network I/O is offloaded to Kotlin Coroutines with a dedicated
Dispatchers.IOpool, preventing UI thread blocking and ensuring a fluid 60FPS typing experience. - Networking Stack: Powered by Retrofit 2 and OkHttp 4, featuring robust connection pooling and timeout management.
- Concurrency Model: Built on Python 3.11+ with
asyncio, enabling non-blocking execution of concurrent AI generation requests. - Multi-Model Resiliency: Features a sophisticated failover mechanism that rotates through multiple Gemini models (1.5 Flash, 2.0 Flash Exp) to handle API-specific constraints or performance degradation.
- Load Balancing & Quota Management: Implements an intelligent key-rotation system that dynamically tracks API quotas and triggers a 24-hour cooldown for exhausted endpoints.
- ⚡ Generative Rewrite: Transforms raw text into professional, casual, polite, or stylized variations in under 800ms.
- 🔍 Contextual Grammar Sanitization: Deep structural analysis of sentences to provide 3 distinct variations of grammatically sound text.
- 🤩 Adaptive UI: A modern, dark-themed interface that responds to user focus and provides immediate AI suggestion feedback.
- 😀 Native Emoji Integration: Seamlessly switches between alpha-numeric, symbol, and emoji layouts without layout jank.
We prioritize a low footprint to ensure Word Tone remains the most reliable input method on any device:
- Memory Footprint: Optimized drawable resources and view recycling keep the application under 50MB of RAM during peak usage.
- Network Efficiency: Responses are compressed and strictly typed in JSON format, minimizing payload size and deserialization time.
- Cold Start Latency: Modular architecture ensures that the IME service initializes only essential components upon system boot.
Word Tone is built with a Privacy-First engineering philosophy:
- Zero Retention: The backend does not persist any user input. Text is processed in-memory and immediately discarded after a successful response.
- Encrypted Transport: All communication between the keyboard and orchestration layer is secured via TLS/SSL.
- Credential Isolation: Sensitive API keys are managed exclusively through environment variables and are never hardcoded or exposed in the client.
- Android Development: JDK 17+, Android SDK 34+.
- Backend Development: Python 3.10+, Virtualenv.
- API Access: A valid Gemini API Key from Google AI Studio.
Developed and maintained by amanjha112113.