AI

Overview FairyField is a desktop AI companion that lives on your desktop as a transparent overlay. Unlike typical AI tools, she’s designed to be a companion — a 3D character with emotions, voice, and long-term memory who remembers your conversations from yesterday. 有灵魂的桌面 AI 伴侣 — 悬浮在桌面上，会笑、会哭、会生气。记得你昨天的烦恼，会用真实的声音跟你聊天。 Philosophy In an era flooded with AI tools, FairyField takes a different path — she’s not a tool, she’s a companion. Not another ChatGPT wrapper Not a cold command-line bot A 3D character with emotions, living on your desktop 100% offline voice processing — your voice data never leaves your machine Long-term memory — she remembers what you said Key Features Module Description 3D Character VRM model with emotion-driven expressions, lip-sync, and eye tracking Offline Voice Paraformer speech recognition + Kokoro speech synthesis + Silero voice detection AI Soul Powered by Claude / GPT with autonomous tool calling and emotional companionship Voice Input Click microphone → speak → auto-transcribe → AI response → voice output Long-Term Memory 4-layer memory stack + temporal knowledge graph + vector semantic search Privacy First Voice 100% offline, memory stored in local SQLite Desktop Overlay Transparent always-on-top window with click-through, non-intrusive Security Prompt injection detection + command whitelist + automatic key sanitization Architecture ┌──────────────────────────────────┐ │ Tauri Transparent Window │ │ ┌──────────┐ ┌───────────────┐ │ │ │ 3D Char │ │ Chat Panel │ │ │ │ Three.js │ │ Vue 3 Bubbles │ │ │ │ + VRM │ │ + Voice Input │ │ │ └──────────┘ └───────────────┘ │ ├──────────────────────────────────┤ │ Tauri IPC Layer │ ├──────────────────────────────────┤ │ Rust Backend │ │ ├── agent/ AI Soul │ │ ├── voice/ ASR/TTS/VAD │ │ ├── memory/ Memory + KG │ │ ├── tools/ Tool System │ │ ├── security/ Safety Guard │ │ ├── gateway/ Discord Notify │ │ └── growth/ Autonomous Growth │ └──────────────────────────────────┘ Tech Stack Layer Technology 3D Rendering Three.js + @pixiv/three-vrm Desktop Framework Tauri 2.0 (Rust) Frontend Vue 3 + TypeScript AI Conversation OpenAI / Anthropic API Speech Recognition sherpa-onnx Paraformer Speech Synthesis sherpa-onnx Kokoro Voice Detection sherpa-onnx Silero VAD Storage SQLite + FTS5 + sqlite-vec Knowledge Graph Temporal Triples Security Prompt Injection Detection + Command Guard + Key Sanitization Project Status Phase Content Status 0 – 4.5 Architecture → Core → Voice → Intelligence → Polish Done 5 MVP Release (real voice engine + voice input UI) Done Future Holographic projection · Data collection · More platforms Planned Testing 341 tests passed (248 Rust + 93 frontend) Source Code Available on GitHub.

Why Multi-Agent? 传统的 AI 对话都是一对一的 — 你问，它答。但现实中的问题往往需要多角度思考。一个技术决策，理性的分析师和创造性的探索者会给出截然不同的建议；一个创意方案，批评者和实用主义者会看到完全不同的盲点。 这个项目的核心问题是：如果我们让多个具有不同人格和专长的 AI 智能体一起讨论，能不能得到比单一 AI 更全面、更有洞见的答案？ 这就是 Pluralistic AI Reasoning — 多元化 AI 推理。 Architecture Decisions Flutter for Everything 选择 Flutter 不是因为它是最流行的，而是因为我们需要真正的跨平台 — Web、iOS、Android、桌面端全覆盖。Multi-Agent Chat 作为一个协作讨论工具，用户可能在手机上快速发起讨论，在桌面上深度参与辩论，或者通过 Web 端分享给没有安装 App 的朋友。 Material 3 的设计语言给了我们统一且现代的 UI 基础，dark/light 主题切换也变得自然。 Firebase as the Invisible Backend 我们选择了 Serverless 架构 — Firebase 提供认证、数据库、实时同步，不需要自己维护任何服务器。Firestore 的 snapshot listener 让多用户实时协作变得简单： // 毫秒级的实时同步就这么简单 firestore.collection('messages').snapshots().listen((snapshot) { // 所有群组成员同时看到新消息 }); Unified SSE Engine 一个关键的技术决策：用一个统一的方法处理所有 LLM 提供商。Kimi、DeepSeek、Qwen、Doubao 都支持 OpenAI 兼容的接口，所以我们的 SSE 流式引擎只需要一个核心方法： ...

Overview Multi-Agent Chat is a cross-platform Flutter application that creates a collaborative AI discussion environment. Instead of interacting with a single AI model, users configure multiple AI agents with distinct personalities and expertise, then watch them debate and collaborate on topics in real-time. The project explores pluralistic AI reasoning — the idea that diverse perspectives from agents with different personas produce richer, more balanced insights than any single model could provide. ...