SumRead

SumRead is a production-minded Android application in active development. It provides a floating overlay that lets the user capture a selected screen region, extract text with on-device OCR, read the text aloud locally, summarize it with an optional AI provider, or open a private in-memory chat grounded in the captured content.

This repository is public yet proprietary, and structured for a serious future Google Play release. The current distribution target is a pre-release APK, not a production Play build.

Current status

• Architecture: MVVM + Clean Architecture + Hilt.
• UI: Jetpack Compose for the app, view-based overlay for system overlay behavior, custom region selection view for capture accuracy.
• Local-first features: overlay control, screen capture flow, region selection, ML Kit OCR, TextToSpeech.
• AI features: Groq, Gemini, and OpenAI behind provider abstractions with user-supplied keys.
• Chat privacy: memory-only session state, no default storage of captured text or chat history.
• Release posture: pre-release APK workflow prepared, production signing still pending.

Feature overview

• Draggable floating overlay button powered by a foreground service.
• User-initiated MediaProjection capture flow with explicit Android consent.
• Region selection before OCR processing.
• On-device ML Kit text recognition.
• On-device text-to-speech with configurable speed, pitch, and language (27 languages supported).
• TTS language pack installer — opens device TTS settings to download additional voices (Chinese, Japanese, Korean, Arabic, Hindi, and more).
• AI summary mode using a selectable provider (Groq / Gemini / OpenAI).
• Smart AI auto-switch — if the selected provider has no API key, the app automatically falls back to another configured provider so AI features keep working.
• AI chat mode grounded in the selected screen content — ask follow-up questions, verify facts, or summarize further without re-capturing.
• Secure local-only storage for provider API keys (see below).
• Compose settings screen with permission helpers and privacy disclosures.

API key storage

API keys are never transmitted or stored in plaintext. Each key is:

1. Encrypted with a hardware-backed AES-256-GCM key generated inside the Android Keystore (never leaves the secure enclave).
2. Stored as a Base64-encoded ciphertext in the app's private SharedPreferences (sumread_secrets).
3. Never included in backups (android:allowBackup="false").
4. Accessible only to this app process — no other app can read them.

To add a key: open Settings → AI providers, enter your key in the relevant field, and tap Save.
To remove a key: tap Clear next to the provider.

Supported providers and where to obtain free or low-cost keys:

Provider	Key source
Groq	console.groq.com — free tier available
Gemini	aistudio.google.com — free tier available
OpenAI	platform.openai.com — pay-as-you-go

Architecture overview

The app follows a single-module clean structure inside app/src/main/java/com/sumread:

• data/ for settings storage, secure secret storage, OCR, TTS, provider integrations, and repository implementations.
• domain/ for models, repository contracts, and use cases.
• presentation/ for Compose screens, activities, viewmodels, and capture/chat UI state.
• service/ for the persistent overlay service and MediaProjection capture service.
• di/ for Hilt modules and network wiring.
• util/ for app configuration, bitmap helpers, intent building, notifications, and geometry helpers.

Tech stack

• Kotlin
• Android SDK 35 target, min SDK 26
• Gradle Kotlin DSL with version catalog
• Jetpack Compose
• Hilt
• DataStore Preferences
• Android Keystore-backed secret storage
• ML Kit Text Recognition
• Android TextToSpeech
• Retrofit + OkHttp
• Groq and Gemini provider integrations

Privacy-first principles

• Local OCR and local TTS are the default path.
• AI features are optional and clearly separated from local-only features.
• The app never ships with embedded provider keys.
• Captured text is not persisted by default.
• Chat history remains memory-only.
• Foreground notifications make long-running sensitive operations visible.
• Screen capture is always user initiated and always gated by the Android system consent dialog.

Download APK

• SumRead.apk — latest build from main
• All releases

Future Google Play release

• Google Play release is planned, not available yet.
• Production signing, store assets, final disclosure copy, and Play review packaging still need completion.
• The repository already includes privacy, data safety, and Play compliance notes to support that path.

Setup

1. Install Android Studio with Android SDK 35 and JDK 17.
2. Clone the private repository.
3. Open the project in Android Studio or use the Gradle wrapper from the repository root.
4. Build with ./gradlew assembleDebug.
5. Install the generated debug APK on a device running Android 8.0 or newer.
6. Open the app, review permissions, and configure optional provider keys in Settings if you want AI features.

Local development

Useful commands:

• ./gradlew testDebugUnitTest
• ./gradlew lintDebug
• ./gradlew assembleDebug

Recommended manual device checks:

• Start and stop the overlay repeatedly.
• Validate MediaProjection consent and cancellation behavior.
• Validate region selection on light and dark backgrounds.
• Validate local read-aloud without any provider key configured.
• Validate AI summary and chat with each provider independently.

Privacy and permissions

Required or optional permissions are intentionally limited:

• SYSTEM_ALERT_WINDOW for the floating overlay.
• FOREGROUND_SERVICE for visible overlay execution.
• FOREGROUND_SERVICE_SPECIAL_USE for the overlay foreground service type (required on Android 14+).
• FOREGROUND_SERVICE_MEDIA_PROJECTION for the active capture service.
• RECORD_AUDIO for optional speech input in chat.
• INTERNET only for optional cloud AI features.

See:

• Privacy policy
• Play Store compliance notes
• Data safety notes

Screenshots

Screenshot placeholders are intentionally reserved until stable device captures are generated:

• docs/screenshots/settings.png
• docs/screenshots/overlay.png
• docs/screenshots/region-selection.png
• docs/screenshots/chat.png

Roadmap

• Add a signed pre-release distribution pipeline.
• Expand OCR language options beyond the current Latin recognizer baseline.
• Add streaming response support behind the existing AI provider abstraction.
• Harden device QA across Android 8 through Android 15.
• Prepare final Play listing assets and store review disclosures.

License

This repository is proprietary and all rights are reserved. See LICENSE.