Private AI on your device.
Download and run local LLMs entirely offline. No cloud. No tracking. No accounts.
Get the latest APK from GitHub Actions → select the latest successful build → download Slate-x.x.x-debug.apk from Artifacts.
Requirements: Android 8.0+, ARM64 device, 4 GB+ RAM
- Run LLMs 100% on-device via llama.cpp — no internet needed after download
- Streaming responses with real-time token display
- Stop / Regenerate / Clear conversation controls
- Conversation starters in 8 languages (EN, DE, FR, ES, TR, AR, ZH, JA)
- Markdown rendering — headers, lists, bold, italic, code, tables
- Syntax highlighting in code blocks (Kotlin, Python, JS, and more)
- Copy button on every message and code block
- Chat history persisted across app restarts (optional, can be disabled)
- 5 built-in models optimized for mobile:
| Model | Size | RAM | Best for |
|---|---|---|---|
| Qwen 2.5 3B | 2.0 GB | 6 GB+ | Best quality, multilingual (recommended) |
| Gemma 3 1B | 806 MB | 3 GB+ | Google's newest, good balance |
| SmolLM2 1.7B | 1.0 GB | 4 GB+ | Simple tasks, fast |
| Phi-3 Mini 3.8B | 2.3 GB | 8 GB+ | Reasoning, coding |
| Llama 3.2 1B | 808 MB | 3 GB+ | Ultra-fast, testing |
- Import your own GGUF models from device storage
- Real downloads with progress, pause/resume, integrity verification
- Device compatibility checks — warns before downloading incompatible models
- Green checkmark on downloaded models
- Zero analytics — no SDKs, no tracking, no telemetry
- Zero accounts — no login, no registration
- Zero cloud — all AI processing on-device
- HTTPS only — encrypted model downloads
- Your controls — delete conversations, models, or all data anytime
- Open source — verify everything yourself
- Dark mode first — anthrazit, ice-blue accents
- Model identity colors — each model has its own accent
- Thinking animation — orbital indicator while AI processes
- Smooth transitions — fade between tabs, slide for details
- Professional typography — clean, readable, minimal
Install the app to see it in action.
# Clone with submodules
git clone --recursive https://github.com/BEKO2210/Android_Engineer.git
cd Android_Engineer
# Build debug APK
./gradlew assembleDebug
# Build release APK (signed with debug key)
./gradlew assembleRelease
# Run tests
./gradlew testDebugUnitTestRequirements: Android SDK 35, NDK 27.2, CMake 3.22.1, JDK 17+
| Layer | Technology |
|---|---|
| Language | Kotlin |
| UI | Jetpack Compose + Material 3 |
| Inference | llama.cpp via JNI (ARM64 NEON) |
| Architecture | MVVM + Clean Architecture (14 modules) |
| DI | Hilt |
| Database | Room |
| Preferences | DataStore |
| Downloads | WorkManager + OkHttp |
| Build | Gradle Kotlin DSL + Version Catalog |
| CI | GitHub Actions |
app/ Application shell, navigation, DI
core/
core-common/ Utilities, device capability, formatting
core-model/ Domain entities
core-data/ Repositories
core-database/ Room DB, DAOs, entities
core-datastore/ DataStore preferences
core-network/ OkHttp client
core-ui/ Design system, theme, shared components
feature/
feature-chat/ Chat screen, streaming, markdown
feature-models/ Model catalog, detail, downloads
feature-settings/ Settings, privacy, storage, import
feature-onboarding/ First-launch flow
inference/
inference-llamacpp/ llama.cpp JNI bridge, native build
download/
download-engine/ WorkManager workers, download manager
See CHANGELOG.md for full release history.
See PRIVACY.md for the complete privacy policy.
TL;DR: Your data never leaves your device. No analytics. No tracking. No cloud. No accounts.
Belkis Aslani — belkis.aslani@gmail.com
App code: See repository license. Model licenses: See MODELS.md for individual model licenses.