moeru-ai/airi
💖🧸 Self hosted, you-owned Grok Companion, a container of souls of waifu, cyber livings to bring them into our worlds, wishing to achieve Neuro-sama's altitude. Capable of realtime voice chat, Minecraft, Factorio playing. Web / macOS / Windows supported.
TypeScript · ★ 46,960 · Fork 4,642 · 在 GitHub 查看
在手机上浏览源码、订阅每日 AI 趋势日报与项目架构深度解读:获取 CopoHub 客户端(iOS & HarmonyOS)→
GitHub Trending 上榜记录
moeru-ai/airi 已在 GitHub Trending 每日榜单上榜 23 天,最高排名第 1,首次上榜 2026-02-27,最近一次上榜 2026-07-30。
| 日期 | 排名 | Star | 当日增长 |
|---|---|---|---|
| 2026-07-30 | #2 | 45,684 | +682 |
| 2026-07-29 | #2 | 45,346 | +797 |
| 2026-07-28 | #3 | 44,714 | +796 |
| 2026-07-27 | #3 | 43,977 | +554 |
| 2026-07-16 | #4 | 42,720 | +110 |
| 2026-07-15 | #4 | 42,464 | +537 |
| 2026-07-14 | #3 | 42,216 | +78 |
| 2026-07-13 | #3 | 41,832 | +57 |
| 2026-05-28 | #17 | 40,423 | +72 |
| 2026-05-27 | #15 | 40,183 | +56 |
| 2026-05-26 | #13 | 39,950 | +62 |
| 2026-05-25 | #13 | 39,673 | +32 |
| 2026-03-30 | #11 | 36,554 | +224 |
| 2026-03-29 | #11 | 36,292 | +213 |
| 2026-03-07 | #1 | 30,024 | +2,562 |
| 2026-03-06 | #1 | 29,439 | +2,544 |
| 2026-03-05 | #5 | 27,225 | +3,003 |
| 2026-03-04 | #11 | 24,558 | +1,794 |
| 2026-03-03 | #3 | 22,104 | +842 |
| 2026-03-02 | #2 | 21,333 | +1,425 |
| 2026-03-01 | #1 | 20,120 | +738 |
| 2026-02-28 | #2 | 19,291 | +1,088 |
| 2026-02-27 | #9 | 18,078 | +229 |
同期上榜的项目
- huggingface/speech-to-speech – Build local voice agents with open-source models
- microsoft/AI-For-Beginners – 12 Weeks, 24 Lessons, AI for All!
- paperswithbacktest/awesome-systematic-trading – A curated list of awesome libraries, packages, strategies, books, blogs, tutorials for systematic trading.
- different-ai/openwork – The open-source alternative to Claude Cowork (powered by opencode)
- WhiskeySockets/Baileys – Socket-based TS/JavaScript API for WhatsApp Web
- pascalorg/editor – Open-source 3D architectural editor with a local CLI, MCP tools, and practical workflows for humans and AI agents.
moeru-ai/airi 项目深度剖析与商业评估报告
moeru-ai/airi 项目深度剖析与商业评估报告
1. 核心定位与商业落地价值
根本痛点:当前 AI 陪伴类产品(Character.ai、SillyTavern 等)局限于纯文本/视觉交互,无法实现"具身化"的数字生命体验。Neuro-sama 虽为标杆,但闭源且无法离线使用。AIRI 的核心价值主张是 "自托管 + 多模态具身交互 + 可编程游戏智能" 三位一体的开源替代方案。
商业落地场景: 虚拟偶像/直播产业:提供完整的 VTuber 技术栈(VRM/Live2D + 实时语音 + 游戏操控),可支撑 7x24 小时无人直播,显著降低 MCN 机构的人力成本。其 Factorio/Minecraft 游戏智能能力已具备 PoC,是差异化卖点。 企业级数字员工/客服:其"容器"架构(Brain/Ears/Mouth/Body 分离)可复用于构建具有 3D 形象的企业级智能助手,嵌入官网或会议系统。 教育/培训模拟:基于游戏环境(Factorio/DomeKeeper)的智能体可作为强化学习与决策 AI 的教学沙盒。
竞品护城河: 跨端覆盖:Web (PWA)、macOS、Windows、iOS 全平台,优于多数仅桌面的竞品(如 semperai/amica 侧重 WebXR)。 Web 原生技术栈:基于 WebGPU/WebAssembly 而非原生引擎,降低了移动端和浏览器的部署门槛,这是架构层面的战略选择。 生态矩阵:已孵化出 unspeech(ASR/TTS 统一网关)、xsai(轻量 LLM SDK)、drizzle-duckdb-wasm(嵌入式数据库驱动)等独立子项目,形成技术护城河。
2. 核心架构设计与技术深度
模块拆解思想:采用 "感知-认知-行动" 的仿生学分层,而非传统的 MVC。核心包 @proj-airi/core 作为事件中枢,通过依赖注入和接口抽象解耦以下模块:
感知层 (Ears):@proj-airi/realtime-audio 处理音频流,接入 Whisper 等 STT,通过 VAD(语音活动检测)实现说话状态机。 认知层 (Brain):通过 xsai 统一接入 30+ LLM Provider(OpenAI、Claude、本地 vLLM/Ollama)。核心设计是 Provider 抽象层,将提示词工程、流式响应、工具调用协议标准化,避免供应商锁定。 行动层 (Mouth/Body):TTS 多 Provider 适配(ElevenLabs、Kokoro 本地推理);VRM/Live2D 动画控制器基于 Three.js 实现,支持自动眨眼、视线追踪等状态机驱动。
在 README 中展示上榜徽章
[](https://github.fq6825.top/repo/moeru-ai/airi)