AI Phone Server

A simple Android app that serves local AI models via an AI with MediaPipe AI inference, object detection, and device sensors for easy integration with an Android terminal.

This is highly experimental, so use it with care and I'm not an Android Developer, so, full disclosure, Google Gemini helped with the UI part.

Download

Download it from the [Releases](https://github.com/parttimenerd/local-a ndroid-ai/releases) page.

Features

AI Inference

• LLM Support: Gemma 3n E2B IT, DeepSeek-R1 Distill Qwen 1.5B, Llama 3.2 (1B/3B), TinyLlama 1.1B
• Object Detection: MediaPipe EfficientDet Lite 2
• Model Management: Download, test, performance metrics (tokens/second)
• Streaming: Real-time token streaming with cancellation support

Device Integration

• Camera: Front/rear with zoom, base64 encoding
• Sensors: Compass orientation (azimuth, pitch, roll)
• Permissions: Runtime request handling

API Endpoints (Port 8005)

• JSON responses with CORS support
• Error handling with HTTP status codes
• Built-in documentation at /help

AI Services

POST /ai/text
{
  "prompt": "text",
  "model": "gemma-3n-e2b-it", 
  "maxTokens": 150,
  "temperature": 0.7,
  "topK": 40,
  "topP": 0.95
}

POST /ai/object_detection
{
  "side": "rear|front",
  "threshold": 0.6,
  "maxResults": 5,
  "returnImage": false
}

GET /ai/models
POST /ai/models/download {"modelName": "model-name"}
POST /ai/models/test {"modelName": "model-name", "prompt": "text"}

Device & System

GET /orientation    # Compass: azimuth, pitch, roll, accuracy  
GET /capture?side=rear&zoom=2.0  # Camera capture, base64 JPEG ⚠️ App must be visible
GET /status         # Server status, features, permissions
GET /help           # API documentation

⚠️ Camera Privacy Notice: Camera capture requires the Android app to be visible due to Android OS privacy restrictions. This ensures users are aware when the camera is being accessed, the other endpoints work in the background.

Build & Install

# Build APK
./gradlew assembleDebug

# Install via ADB
adb install -r app/build/outputs/apk/debug/app-debug.apk

# Monitor logs
adb logcat -s "LocalAIPhoneServer" "*AI*"

Integration

Designed for K3s cluster nodes running on Android devices. Provides REST API for cluster applications to access AI inference, device sensors, and camera functionality.

GET /health
# Simple health check endpoint

GET /help
# Complete API documentation with examples

🛠️ Build Instructions

Prerequisites

• Android Studio Arctic Fox or newer
• Android SDK API level 24+ (Android 7.0+)
• Device Requirements: 3GB+ RAM for AI features
• Permissions: Camera

Building the App

# Clone the repository
git clone <repository-url>
cd android

# Build debug APK
./gradlew assembleDebug

# Install on connected device  
./gradlew installDebug

# Build release APK
./gradlew assembleRelease

Debug Information

The /status endpoint provides comprehensive debug information including:

• Current AI model status and memory usage
• Permission status for all features
• Available device memory and requirements
• Feature availability and configuration

TODO

• Location isn't working

License

Apache 2.0