Orion

An Android AI assistant that runs models locally on-device and connects to cloud providers when needed. Built with Kotlin, Jetpack Compose, and llama.cpp.

Features

🤖 Multiple AI Modes

• Local: Run GGUF models entirely on-device using llama.cpp
• Cloud: Connect to GPT-4, Claude, Gemini, DeepSeek, Qwen, Kimi
• Live: Real-time voice conversation via OpenAI Realtime API
• Emergency: Mesh networking for offline communication via WiFi Direct

🔐 Privacy & Security

• No hardcoded API keys - users provide their own
• Encrypted conversation memory
• Local-first architecture
• No analytics or telemetry
• Open source

🆘 Emergency Mode

WiFi Direct mesh network for communication without internet:

• E2E encryption (ECDH + AES-256-GCM)
• Multi-hop message routing
• Store-and-forward protocol
• QR code contact sharing
• Works completely offline

Technical Stack

• Language: Kotlin
• UI: Jetpack Compose
• Local LLM: llama.cpp (JNI integration)
• Networking: OkHttp, WiFi Direct P2P
• Build: Gradle with NDK support

Architecture

app/src/main/java/com/orion/proyectoorion/
├── ai/                 # Local LLM engine and llama.cpp bridge
├── data/               # Data persistence and encrypted memory
├── emergency/          # Mesh networking and emergency communications
├── models/             # Cloud provider integrations and model definitions
├── ui/                 # Jetpack Compose screens and components
├── vivo/               # Real-time voice conversation engine
└── viewmodel/          # App logic and state management

Build Instructions

Prerequisites

• Android Studio Hedgehog or higher
• NDK 27.0.12077973
• CMake 3.22.1
• JDK 11+

Build Commands

./gradlew assembleDebug    # Debug build
./gradlew assembleRelease  # Release build
./gradlew bundleRelease    # Play Store bundle

Requirements

• Minimum SDK: Android 11 (API 30)
• Architecture: ARM64
• Permissions: Camera, WiFi, Location (for WiFi Direct)

Development Philosophy

This project uses numbered version snapshots (v1, v2, v3...) with descriptive suffixes:

• funciona - in development
• perfecto - stable
• errores - has bugs (for debugging)
• estable - release candidate
• produccion - production ready

This approach allows quick rollbacks during rapid C++ integration development.

Version v99

This version represents a stable foundation with:

• Working llama.cpp integration
• All cloud providers operational
• Functional memory system
• Complete basic UI
• Emergency mesh networking

Third-Party Libraries

• llama.cpp - Local LLM inference (MIT)
• ZXing - QR code generation/scanning
• OkHttp - HTTP client
• Jetpack Compose - UI framework

License

[Add license here]

Related Projects

• Genesis Simulation - AI personality systems
• Are You There Reading? - Quantum randomness exploration
• PCP Universe - Information persistence theory
• Physics Discovery AI - Multi-agent physics discovery