airi

Project AIRI is an open-source, self-hosted AI companion project that aims to recreate the experience of Neuro-sama: a virtual character or "cyber living" you own and can interact with anytime, anywhere. It is positioned as a container of souls for AI waifu and virtual characters, bringing them into real-time interaction with users. Core capabilities evidenced in the README: - Realtime voice chat: audio input from the browser or Discord, client-side speech recognition, and talking detection. Voice synthesis supports multiple providers including ElevenLabs, Microsoft/Azure Speech, OpenAI-compatible TTS, Alibaba Cloud Model Studio, and local Kokoro TTS. - Game playing: the project lists Minecraft and Factorio as playable (Factorio is work-in-progress with a proof of concept), plus Kerbal Space Program announced, and Helldivers 2 co-play as WIP. - Chat integrations: Telegram and Discord chat are supported. - Avatars: VRM and Live2D model support with animations such as auto blink, auto look-at, and idle eye movement. - Memory: pure in-browser database support via DuckDB WASM and pglite, with a memory system in progress. - Platform support: browser (PWA), macOS, Windows, and a mobile version (Stage Pocket, experimental). Desktop builds can use native NVIDIA CUDA and Apple Metal through the candle project. - LLM provider support: a long list of providers including OpenAI, Azure OpenAI, Anthropic Claude, DeepSeek, Qwen, Google Gemini, xAI, Groq, Mistral, Cloudflare Workers AI, Together.ai, Fireworks.ai, Novita, Zhipu, SiliconFlow, Stepfun, Baichuan, Minimax, Moonshot AI, ModelScope, Player2, Tencent Cloud, Xiaomi Mimo, Ollama, vLLM, SGLang, OpenRouter, and AIHubMix. The project is built with web technologies such as WebGPU, WebAudio, Web Workers, WebAssembly, and WebSocket, and is organized as a monorepo with stage applications (web, desktop, pocket), shared packages, a desktop server channel, integrations (Discord bot, Minecraft bot, Telegram, Factorio, MCP), and a hosted backend using PostgreSQL and Redis. Installation options include winget, Scoop, Homebrew Cask, and Nix. Development uses pnpm with separate commands for the browser, desktop, mobile, and documentation stages. The README also notes a translation project on Crowdin and a warning that there is no official cryptocurrency or token associated with the project. Sub-projects include unspeech (ASR/TTS proxy), hfup, xsai-transformers, WebAI realtime voice chat example, DuckDB WASM packages, AIRI Factorio, AIRI DomeKeeper, Factorio RCON API, autorio, Velin, demodel, inventory, MCP Launcher, and SAD documentation for self-hosted and browser LLMs.