AIRI (Project AIRI) is an open-source AI companion and virtual character, modeled on the AI VTuber Neuro-sama. Beyond chat, the character can play games, listen and speak in real time, and appear as an animated Live2D or VRM avatar. It is aimed at people who want to own and run their own digital companion rather than use a closed character platform.
The project is built on Web technologies (WebGPU, WebAudio, Web Workers, WebAssembly, and WebSocket), so the character runs in a modern browser and on mobile as a PWA. A desktop version (Stage Tamagotchi) uses native NVIDIA CUDA and Apple Metal through the Hugging Face candle library, and adds features that need non-Web connections, such as joining a Discord voice channel or playing Minecraft and Factorio. A mobile version (Stage Pocket) is experimental.
LLM access goes through the project's xsai library, which supports a long list of API providers including OpenAI, Azure OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Qwen, xAI, Mistral, Zhipu, Minimax, Moonshot AI, Xiaomi Mimo, OpenRouter, Groq, and Cloudflare Workers AI, plus self-hosted servers such as Ollama, vLLM, and SGLang. Memory uses an in-browser database (DuckDB WASM or pglite), with a longer-term memory system in progress.
Features
- Game playing: plays Minecraft, Factorio, and Kerbal Space Program, with Helldivers 2 co-play in progress
- Chat channels: talks in Telegram and Discord, including audio input from Discord voice
- Speech input: browser audio input with client-side speech recognition and talking detection
- Voice synthesis: multi-provider text-to-speech including ElevenLabs, Microsoft/Azure Speech, OpenAI-compatible TTS, Alibaba Cloud Model Studio, and local Kokoro TTS
- Avatars: VRM and Live2D models with automatic blinking, look-at, and idle eye movement
- Cross-platform stages: browser (Stage Web), desktop for Windows, macOS, and Linux (Stage Tamagotchi), and an experimental mobile app (Stage Pocket)
- In-browser memory: DuckDB WASM or pglite as an embedded database
- Broad provider support: cloud LLM APIs and self-hosted inference servers through xsai
