Airi — Self-Hosted AI Companion with Real-Time Voice Chat

Introduction

Airi is an open-source project that lets you run a self-hosted AI companion capable of real-time voice chat, game integration (Minecraft, Factorio), and expressive avatar rendering via Live2D and VRM models. It aims to give users full ownership of their AI companion without relying on third-party services.

What Airi Does

• Provides real-time voice conversation with configurable TTS and STT backends
• Renders interactive Live2D and VRM avatars in a browser-based interface
• Integrates with games like Minecraft and Factorio for in-game AI interaction
• Supports multiple LLM providers for personality and dialogue generation
• Runs entirely self-hosted on your own hardware for complete data privacy

Architecture Overview

Airi is built as a monorepo using TypeScript and pnpm workspaces. The frontend renders avatars using PixiJS for Live2D and Three.js for VRM models. Voice pipelines handle STT, LLM inference, and TTS in a streaming loop. Game integrations use dedicated bot modules that connect to game servers via their respective protocols. The system is modular so each component (voice, avatar, game) can be enabled or disabled independently.

Self-Hosting & Configuration

• Clone the repository and install dependencies with pnpm
• Configure your LLM provider credentials in the environment file
• Set up TTS and STT endpoints or use built-in browser Web Speech API
• Optionally enable game bot modules for Minecraft or Factorio
• Deploy with Docker or run directly with Node.js on any modern OS

Key Features

• Full self-hosted deployment with no external service dependency
• Cross-platform support for web browsers, macOS, and Windows
• Modular architecture allowing selective feature activation
• Extensible avatar system supporting both Live2D and VRM formats
• Active community with regular updates and new integrations

Comparison with Similar Tools

• SillyTavern — focused on text chat and character cards; Airi adds voice and game integration
• Vtuber Studio — avatar rendering only without conversational AI capabilities
• GPT4All — local LLM inference without avatar or voice features
• Jan — offline chat client; Airi extends into gaming and visual avatars
• Neuro-sama — proprietary; Airi provides an open-source alternative

FAQ

Q: What hardware do I need to run Airi? A: A modern desktop with Node.js 18+ is sufficient. GPU is optional but recommended for local LLM inference.

Q: Can I use my own LLM models? A: Yes. Airi supports OpenAI-compatible endpoints, Ollama, and other configurable providers.

Q: Does Airi work without an internet connection? A: It can if you use a local LLM and local TTS/STT. Game integrations require network access to the game server.

Q: Is Airi suitable for production use? A: It is primarily a community project for personal use and experimentation, not a production-grade product.

Sources

• https://github.com/moeru-ai/airi
• https://airi.moeru.dev