AIRI is an open-source, locally-run, customizable virtual AI assistant developed by Moeru AI. It aims to provide a privacy-focused alternative to cloud-based AI assistants like ChatGPT, offering users complete control over their data and the AI's behavior. The core of AIRI is built around a modular architecture, allowing for extensive customization and extension through plugins and different language models. It's designed to be accessible to both developers and end-users, with a focus on ease of use and a growing community.

The project distinguishes itself through its emphasis on "personality" and emotional responsiveness. Unlike many AI assistants focused solely on task completion, AIRI strives to create a more engaging and believable interaction experience. This is achieved through a combination of carefully crafted prompts, a dedicated "memory" system, and the ability to define a detailed character profile for the AI. Users can specify AIRI's backstory, preferences, and even emotional tendencies, influencing how it responds to prompts and interacts over time. This personality is not just superficial; it's intended to be integrated into the AI's reasoning and decision-making processes.

Technically, AIRI leverages Large Language Models (LLMs) but doesn't require a constant internet connection after initial setup. It supports various LLMs, including those from the Llama.cpp ecosystem (like Llama 2, Mistral, and others) and OpenAI's models (requiring an API key). The project provides tools for easily switching between models, allowing users to experiment and find the best balance between performance, cost, and privacy. A key component is the "Airi Core," which handles the communication with the LLM and manages the AI's state. The core is written in Python and utilizes a web-based interface built with Gradio, making it accessible through a web browser.

The plugin system is a significant feature, enabling users to extend AIRI's functionality beyond basic conversation. Plugins can add capabilities like web browsing, image generation, code execution, and integration with external services. The repository includes examples of several plugins, and the framework is designed to encourage community contributions. This modularity allows AIRI to adapt to a wide range of use cases, from personal assistance and creative writing to research and automation. The plugin architecture is designed to be relatively simple, making it easier for developers to create and share new functionalities.

Currently, the project is under active development, with frequent updates and improvements. The GitHub repository contains comprehensive documentation, including installation instructions, API references, and guides for creating plugins and customizing AIRI's personality. The community is active on Discord, providing support and sharing their creations. While still evolving, AIRI represents a promising step towards more personalized, private, and powerful local AI assistants, offering a compelling alternative to cloud-based solutions and empowering users to take control of their AI experience.