Llama Compose is a showcase app for Colombia AI Week, designed to highlight on-device AI experiences with Android and Google technologies. Built with Kotlin Multiplatform and optimized for Android, it demonstrates how advanced AI models can run locally on user devices, enabling interactive conversations without relying on cloud processing. The app supports both simple and agent-based chat modes and allows users to download and manage models directly on their phones.
Key features:
- On-device AI inference using llama.cpp
- Support for Google’s Gemma and Meta’s Llama models
- Multiple conversation modes (Simple & Agent)
- Agent functionality with tool calling through Koog.ai
- Local model download, storage, and management
- Built with Kotlin Multiplatform, optimized for Android
- Real-time, interactive chat experience powered entirely on-device
Important Disclaimer: This app includes experimental AI functionality. Model outputs may be offensive, inaccurate, or inappropriate. Users should exercise caution and avoid depending on this app for sensitive or critical decisions. It is intended for educational and demonstration purposes only.