Summary

IRIS is an offline Android chat application enabling local interfacing with GGUF/llama.cpp models. It targets users seeking private, secure, and fully offline AI interactions, allowing direct on-device LLM execution. The app offers customizable parameters and expandable model support for a flexible, user-controlled AI experience.

How It Works

A fork of the llama.cpp Android example, IRIS uses its core architecture for on-device LLM inference. It supports downloading GGUF models from Hugging Face, with direct in-app management. Key design choices include a privacy-first, offline-only approach, and customizable inference parameters (n_threads, top_k, top_p, temperature) to optimize performance based on device capabilities. Text-to-Speech and Speech-to-Text are integrated.

Quick Start & Requirements

Install via Google Play or GitHub releases. For building from source, clone iris_android and llama.cpp, then checkout commit 1f922254f0c984a8fb9fbaa0c390d7ffae49aedb in llama.cpp. Requires Android Studio and developer options enabled on the device. Performance depends heavily on the chosen GGUF model size and device hardware.

Highlighted Details

• Offline & Private: All AI processing is local, ensuring data privacy and offline functionality.
• Model Flexibility: Supports downloading and running various GGUF models from Hugging Face.
• Customizable Inference: Tune parameters like n_threads, top_k, top_p, and temperature.
• Multimedia Integration: Includes Text-to-Speech and Speech-to-Text.

Maintenance & Community

Maintained by Nerve Sparks (www.nervesparks.com). Contributions follow standard GitHub PRs. No specific community channels or roadmaps are detailed in the README.

Licensing & Compatibility

The software license is not explicitly stated in the README. Compatibility is for Android devices. Clarification is needed for commercial use or integration.

Limitations & Caveats

AI response accuracy and performance depend on model complexity and device resources, potentially impacting results for complex queries. Building from source requires a specific, older llama.cpp commit, suggesting potential dependency management issues. The project's development stage (e.g., alpha/beta) is not specified.