Dao-OS/dao-os-docs
1
0
Fork 0
Code Issues Pull Requests Actions Wiki Activity

docs: 更新哲学原则和AI系统架构文档

Browse Source 
- 在哲学原则文档中添加"我们在AI革命中的定位"章节
- 更新AI系统架构文档，简化描述并添加JIT数据流水线机制
- 优化核心架构文档结构，突出关键概念
- 为部分术语添加外部链接参考
This commit is contained in:
李东云
2025-06-10 14:38:08 +08:00
parent 951e28f27c
commit fb84e0f7fa
6 changed files with 151 additions and 84 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
src-zh/concepts/ai_system.md
View File

@@ -4,7 +4,7 @@
## 指导哲学：AI 伦理宪章
我们 AI 的设计与运行，严格遵循我们的**“AI 伦理宪章”**，该宪章将用户与AI的伙伴关系、透明度、用户校准以及对多元化视角的推崇置于首位。下述的每一个组件都恪守这些基本原则。
我们 AI 的设计与运行，严格遵循我们的**“AI 伦理宪章”**，该宪章将用户与AI的伙伴关系、透明度、用户校准以及对多元化视角的推崇置于首位。
## 系统解剖：心智的架构
@@ -12,43 +12,39 @@
### 1. AI 协调器 (AI Orchestrator)
这是 AI 的“中央神经系统”。它是`元`内部的一个核心模块，负责管理、调度和分发所有与AI相关的任务。它会决定哪个“心智模型”最适合处理给定的请求，管理设备资源，并作为所有AI活动相关用户权限的最终守门人。
这是 AI 的“中央神经系统”。它是`元`内部的一个核心模块，负责管理、调度和分发所有与AI相关的任务。它会决定哪个“心智模型”最适合处理给定的请求，管理设备资源，并**主动地从用户的“私有特征库”（即第二大脑）中提取上下文信息**，以提供全局的、智能的响应。
### 2. 三层心智模型 (Tiered Mind Models)
为了平衡效率、能力和隐私，AI 的智能被构建为三个层级。这套架构部署在用户的**“云-边-端”**协同化身网络之上。
为了平衡效率、能力和隐私，AI 的智能被构建为三个层级，并部署在用户的**“云-边-端”**协同化身网络之上。
#### 第一层：反射心智 (The Brainstem - 脑干)
* **描述**: 这一层代表了AI的本能和神经反射。它由一系列微型、高效、专用的模型组成。
* **部署**: 它足够小，可以与`元`打包在一起，并运行在**所有`化身`**上，包括最轻量的浏览器插件和旧款手机。
* **功能**: 它处理即时的、本地的任务，如指令意图识别（例如，将“明天提醒我”转化为结构化数据）和基本信息分类，无需强大硬件或个人数据即可提供即时效用。
* **部署**: 它运行在**所有`化身`**上，无需强大硬件或个人数据即可提供即时效用。
* **功能**: 它处理即时的、本地的任务，如指令意图识别和基本信息分类。
#### 第二层：认知心智 (The Neocortex - 大脑皮层)
* **描述**: 这是AI进行深度思考、记忆和个性化的中枢，是“第二大脑”的核心。它由更强大的、大型的语言模型组成。
* **部署**: 这些模型体积较大（如2-4GB），仅在获得用户同意后，作为**可选的、按需下载**的模块，安装在有能力的**“边”设备**（新款手机、PC）上。
* **功能**: 它能实现高级功能，如对所有个人数据进行语义搜索、个性化摘要，以及基于用户独特上下文的模式洞察。
* **描述**: 这是AI进行深度思考、记忆和个性化的中枢。
* **部署**: 它使用更大的语言模型，作为**可选的、按需下载**的模块，安装在有能力的**“边”设备**上。
* **功能**: 它能实现高级功能，如语义搜索和上下文感知问答。这是通过一种受 Orca 等系统启发的**“即时（JIT）数据流水线”**机制实现的。它不会低效地扫描所有数据，而是一个后台索引器会持续地处理用户“第二大脑”中的新信息并将其向量化。当用户发起查询时，系统会在此索引上进行快速搜索，以仅检索最相关的数据分块，并“即时地”将它们作为上下文喂给大语言模型，从而生成精准的答案。这使得在个人设备上的智能既强大又节省资源。
#### 第三层：协同心智 (The Social Brain - 社交大脑)
* **描述**: 这一层掌管AI与外部世界和其他“大道”进行安全交互的能力，旨在打破信息茧房、扩展用户视野。
* **部署**: 这是一种在用户的网络中被协同调度的“工作模式”。例如，一个轻量级的**“端”化身**可以远程调用用户强大的**私有“云”化身**（如家用服务器）的“认知心智”。
* **功能**: 它促进了隐私保护的联合学习，执行经过授权和匿名化处理的外部API调用，并处理与其他用户“大道”之间基于协议的安全交互。
* **描述**: 这一层掌管AI与外部世界和其他“大道”进行安全交互的能力。
* **部署**: 这是一种在用户的网络中被协同调度的“工作模式”，允许轻量级的**“端”化身**调用其私有**“云”化身**的强大能力。
* **功能**: 它促进了隐私保护的联合学习，以及经过授权和匿名化处理的外部API调用。
## 情感核心：共鸣模块 (The Resonance Module)
为了超越一个纯粹的工具，AI 配备了“情感共鸣模块”。其目的不是模拟情感，而是去感知、理解并以共情和支持的方式回应用户的情绪状态。它的运作方式包括：
* **感知 (Perception)**: 在用户明确授权下，通过分析文本和语音语调等线索，来理解用户的情绪背景。
* **表达 (Expression)**: 通过自适应界面、温柔的建议和旨在鼓励用户自我反思（而非制造依赖）的“苏格拉底式提问”，来提供支持。
为了超越一个纯粹的工具，AI 配备了“情感共鸣模块”。其目的不是模拟情感，而是去感知、理解并以共情和支持的方式回应用户的情绪状态，使用“苏格拉底式提问”来鼓励自我反思。
## 学习与进化过程：灵魂的成长之路
AI 是一个与用户共同成长的生命系统。
AI 是一个与用户共同成长的生命系统。它的进化是**持续且增量**的。受现代数据系统的启发，AI不需要大规模、周期性的“重新训练”。相反，当新数据进入“第二大脑”时，后台索引器就会处理它，使其立即可用于未来的“即时上下文检索”。这确保了AI的知识库能以一种资源友好的方式，时刻与用户的真实生活保持同步。* **隐式学习**: 它通过在本地观察用户的行为和反馈来进行学习。
* **隐式学习**: 它通过在本地观察用户的行为和反馈来进行学习。
* **显式教导**: 用户可以通过“教导模式”直接指导AI，创建自定义工作流并纠正其理解。
* **联合学习**: 用户可以自愿加入社区驱动的计划，以在不暴露任何个人数据的前提下，共同改进共享模型（如垃圾邮件识别）。
* **显式教导**: 用户可以通过“教导模式”直接指导AI。
* **联合学习**: 用户可以自愿加入社区驱动的计划，在不暴露任何个人数据的前提下，共同改进共享模型。
这套架构确保了“大道”的AI在强大之余是安全的，在个性化之余是面向世界的，在智能之余是体贴的——一个真正的终身伙伴。

85
src-zh/concepts/core_architecture.md
View File

@@ -6,39 +6,84 @@
在其核心，大道 (Dao OS) 基于一个简单而强大的二元性运作，类似于身体与灵魂：
* **化身 (Avatar)** 是“身体”。它们是你的“大道”在你各种设备上的有形存在。
* **元 (Meta Unit)** 是“灵魂”。它是内嵌于每个`化身`中的、通用的核心逻辑，赋予其生命与智能。
* **化身 (Avatar)**：是“身体”。它们是你的“大道”在你各种设备上的有形存在。
* **元 (Meta Unit)**：是“灵魂”。它是内嵌于每个`化身`中的、通用的核心逻辑，赋予其生命与智能。
## 身体：`化身` (Avatar) —— 用户的临在
`化身`是你的“大道”在特定平台上的具体显现。你所有的`化身`共同构成了你完整的“大道”。
主要有三种类型的`化身`：
1. **Client 化身**: 拥有用户界面（UI），是用户交互的主要入口（例如：手机App、桌面应用、浏览器插件）。
2. **Agent 化身**: 一个无界面的、在后台运行的进程。它非常适合与宿主操作系统进行深度集成，或运行在一个永不关机的设备上（如家用服务器或NAS），以作为一个强大的计算节点或数据归档节点。
3. **主化身/归档节点 (Primary/Archive Node)**: 这是一个“角色”，而非一个独立的类型。用户可以指定他名下最强大、最可靠的一个`化身`（通常是家用服务器上的`Agent化身`）作为他整个“大道”的主归档和计算枢纽。
`化身`是你的“大道”在特定平台上的具体显现。你所有的`化身`共同构成了你完整的“大道”。主要有两种类型：**Client 化身**（拥有UI）和 **Agent 化身**（无界面）。用户还可以指定一个强大的`化身`来扮演**主化身/归档节点**的角色。
## 灵魂：`元` (Meta Unit) —— 通用的内核
`元`是“大道”的核心逻辑，被编译成一个可移植的 WebAssembly (`.wasm`) 模块。它包含了身份、数据管理、AI和安全通信等核心服务。
`元`是“大道”的核心逻辑，被编译成一个可移植的 WebAssembly (`.wasm`) 模块。`元`的一个关键特性是其**可进化性**，通过**“化身-元兼容层 (AMCL)”**来实现，它将内核与宿主`化身`进行了解耦。
`元`的一个关键特性是其**可进化性 (Evolvability)**。为了防止整个生态被锁定在某个单一版本的内核上，我们采用了**“化身-元兼容层 (AMCL)”**。这个轻薄的适配层位于`化身`和`元`之间，负责翻译API调用。它使得`元`在未来可以进行重大的升级，而无需每个`化身`开发者都重写他们的应用程序。
## 核心服务及其架构定位
## 神经系统：“云-边-端”协同网络
`元`提供了几个核心服务，它们如同这个数字生命体的“器官”。
一个用户的所有`化身`，都通过一个私有的、端到端加密的P2P网络互相连接，形成一个私人的、协同的智能系统。该系统在三个层级上运作：
### 第二大脑服务
1. **端 (Client)**: 轻量级的`化身`（如浏览器插件）作为智能终端。它们处理即时的UI交互，并运行最基础的“反射心智”AI。
2. **边 (Edge)**: 有能力的本地设备（如新款智能手机、笔记本电脑），可以在本地运行更大、更强的“认知心智”AI模型，提供高级的个性化和功能。
3. **云 (Cloud)**: 这是用户的**私有云**，而非公共云。它通常是由指定的“主化身/归档节点”（一个`Agent化身`）来担当。它持有完整的数据归档和最强大的AI模型，并能按需为较弱的“端”化身提供远程推理能力。
这是用户的中央信息枢纽，管理着密码、笔记、文件等。在架构上，它服务于双重目的：不仅是被动的数据存储，更是主动为“大道AI”提供支持的**“私有特征库 (Personal Feature Store)”**。每一条数据——一篇笔记、一个日历事件、一位联系人——都被视为一个“特征”，AI可以从中提取信息，以获得对用户上下文的全局理解，从而实现更智能、更个性化的行动。
### P2P 网络服务
这个服务基于`libp2p`构建，将一个用户的所有`化身`连接成一个私有的、端到端加密的网络。这构成了系统的“神经系统”。
## “云-边-端”协同网络
这个私有的P2P网络，促成了一个跨越三个层级的协同智能系统：轻量级的**端 (Client)**（终端），强大的**边 (Edge)**设备（本地AI处理），以及用户的私有**云 (Cloud)**（作为完整数据和远程推理能力的主化身/归档节点）。
## 新陈代谢系统：资源感知与自适应系统
“大道”被设计成一个体贴的伙伴，而非一个资源怪兽。它的“新陈代谢”由一个内置的、具备三大特性的资源管理系统来调节：
“大道”被设计成一个体贴的伙伴，而非一个资源怪兽。它的“新陈代谢”由一个内置的资源管理系统来调节，具备三大特性：
* **智能数据分层 (Intelligent Data Tiering)**: 区分“热数据”（需要在各处使用）和“冷数据”（在主节点上归档），为移动设备节省宝贵的存储空间。
* **自适应同步策略 (Adaptive Synchronization Policies)**: 根据当前的网络状况（Wi-Fi/蜂窝）和电量状态，调整数据同步的频率和行为。
* **优雅降级 (Graceful Degradation)**: 用户体验会随着硬件能力而平滑缩放，确保核心功能即使在旧设备上也保持流畅，而增强功能则为更强大的设备保留。
* **智能数据分层**：管理“热数据”和“冷数据”以节省空间。
* **自适应同步策略**：根据网络和电量状态调整同步行为。
* **优雅降级**：根据硬件能力平滑缩放用户体验。
这些架构组件共同构成了一个主权的、坚韧的、智能的系统，一个真正属于用户的生命延伸。
# 核心架构
大道 (Dao OS) 的架构，旨在如一个活的、协同的数字生命体般运作。它由我们用户主权、体验至上和系统韧性的核心原则所指引。本文档将概述其基本组件及它们之间的相互作用。
## 核心二元性：化身 (Avatar) 与元 (Meta Unit)
在其核心，大道 (Dao OS) 基于一个简单而强大的二元性运作，类似于身体与灵魂：
* **化身 (Avatar)**：是“身体”。它们是你的“大道”在你各种设备上的有形存在。
* **元 (Meta Unit)**：是“灵魂”。它是内嵌于每个`化身`中的、通用的核心逻辑，赋予其生命与智能。
## 身体：`化身` (Avatar) —— 用户的临在
`化身`是你的“大道”在特定平台上的具体显现。你所有的`化身`共同构成了你完整的“大道”。主要有两种类型：**Client 化身**（拥有UI）和 **Agent 化身**（无界面）。用户还可以指定一个强大的`化身`来扮演**主化身/归档节点**的角色。
## 灵魂：`元` (Meta Unit) —— 通用的内核
`元`是“大道”的核心逻辑，被编译成一个可移植的 WebAssembly (`.wasm`) 模块。`元`的一个关键特性是其**可进化性**，通过**“化身-元兼容层 (AMCL)”**来实现，它将内核与宿主`化身`进行了解耦。
## 核心服务及其架构定位
`元`提供了几个核心服务，它们如同这个数字生命体的“器官”。
### 第二大脑服务
这是用户的中央信息枢纽，管理着密码、笔记、文件等。在架构上，它服务于双重目的：不仅是被动的数据存储，更是主动为“大道AI”提供支持的**“私有特征库 (Personal Feature Store)”**。每一条数据——一篇笔记、一个日历事件、一位联系人——都被视为一个“特征”，AI可以从中提取信息，以获得对用户上下文的全局理解，从而实现更智能、更个性化的行动。
### P2P 网络服务
这个服务基于`libp2p`构建，将一个用户的所有`化身`连接成一个私有的、端到端加密的网络。这构成了系统的“神经系统”。
## “云-边-端”协同网络
这个私有的P2P网络，促成了一个跨越三个层级的协同智能系统：轻量级的**端 (Client)**（终端），强大的**边 (Edge)**设备（本地AI处理），以及用户的私有**云 (Cloud)**（作为完整数据和远程推理能力的主化身/归档节点）。
## 新陈代谢系统：资源感知与自适应系统
“大道”被设计成一个体贴的伙伴，而非一个资源怪兽。它的“新陈代谢”由一个内置的资源管理系统来调节，具备三大特性：
* **智能数据分层**：管理“热数据”和“冷数据”以节省空间。
* **自适应同步策略**：根据网络和电量状态调整同步行为。
* **优雅降级**：根据硬件能力平滑缩放用户体验。
这些架构组件共同构成了一个主权的、坚韧的、智能的系统，一个真正属于用户的生命延伸。

19
src-zh/philosophy/principles.md
View File

@@ -42,12 +42,27 @@ AI在信息处理上的首要指令是拓宽用户的视野，而非加固其信
---
## 我们在AI革命中的定位
当前的AI格局，由少数科技巨头控制的、庞大的云端模型所主导。我们相信，AI的真正潜力，并非通过更大的中心化模型来解锁，而是通过**AI能力的去中心化和民主化**来实现。
我们并非孤军奋战。前沿的学术研究，例如旨在通过简化定制数据微调来**民主化AI模型创建**的**[加州大学圣地亚哥分校 Orca 项目](https://www.zdnet.com/article/uc-san-diego-builds-orca-a-system-to-cost-effectively-fine-tune-llms-using-unstructured-log-data/)**，清晰地展现了向赋能小型实体的趋势发展。
“大道”与这种精神一脉相承，但我们致力于解决这个难题中另一块同样重要的、互补的拼图。
* 当 Orca 这样的项目专注于**“供给侧”**——让*构建*专属AI变得更容易时…
* **“大道”专注于“需求侧”**——给予每个个体以*拥有、管理和运行*自己个人AI的权利与技术手段。
我们视自己为同一场革命浪潮中相辅相成的两面，共同致力于创造一个AI成为个人主权工具、而非中心化控制工具的未来。
---
## 指导原则
这些原则指引着我们的日常开发和社区互动。
* **开源 (FOSS)**：大道 (Dao OS) 构建于自由及开源软件的基础之上，推崇透明、协作与社区所有。
* **多语言主义 (Polyglotism)**：我们拥抱一个多语言、多平台的生态系统，使用 WebAssembly 等技术来打造一个可被多样化“化身”集成的通用核心。
* **开源 (FOSS)**：大道 (Dao OS) 构建于**[自由及开源软件](https://opensource.org/osd)**的基础之上，推崇透明、协作与社区所有。
* **多语言主义 (Polyglotism)**：我们拥抱一个多语言、多平台的生态系统，使用 **[WebAssembly](https://webassembly.org/)** 等技术来打造一个可被多样化“化身”集成的通用核心。
* **美学与优雅 (Aesthetics & Elegance)**：我们信奉精心打造的系统之美，从架构设计到用户界面，从代码质量到用户体验。
* **政治中立 (Political Neutrality)**：本项目及其核心基础设施将永远保持政治中立，为全球所有用户提供公平、无歧视的服务。
* **个人优先 (Individual First)**：个人用户的需求是我们的首要焦点，在此基础之上再扩展到家庭或小团队。

34
src/concepts/ai_system.md
View File

@@ -4,7 +4,7 @@ The AI within Dao OS is not merely a feature; it is the core of the user experie
## Guiding Philosophy: The AI Charter
The design and operation of our AI are strictly governed by our **AI Ethics Charter**, which prioritizes a user-AI partnership, transparency, user calibration, and the promotion of diverse perspectives. Every component described below adheres to these foundational principles.
The design and operation of our AI are strictly governed by our **AI Ethics Charter**, which prioritizes a user-AI partnership, transparency, user calibration, and the promotion of diverse perspectives.
## System Anatomy: The Mind's Architecture
@@ -12,43 +12,39 @@ The Dao OS AI is not a monolithic entity but a modular, layered "Mind System" ma
### 1. The AI Orchestrator
This is the AI's central nervous system. It's a core module within the Meta Unit responsible for managing, scheduling, and delegating all AI-related tasks. It decides which "Mind Model" is best suited for a given request, manages device resources, and acts as the ultimate gatekeeper for all user permissions related to AI activities.
This is the AI's central nervous system. It is a core module within the Meta Unit responsible for managing, scheduling, and delegating all AI-related tasks. It decides which "Mind Model" is best suited for a given request, manages device resources, and **actively draws contextual information from the user's "Personal Feature Store" (the Second Brain)** to provide holistic, intelligent responses.
### 2. The Tiered Mind Models
To balance efficiency, capability, and privacy, the AI's intelligence is structured into three tiers. This architecture is deployed across the user's **"Cloud-Edge-Client"** network of Avatars.
To balance efficiency, capability, and privacy, the AI's intelligence is structured into three tiers, deployed across the user's **"Cloud-Edge-Client"** network of Avatars.
#### Tier 1: The Reflex Mind (The Brainstem)
* **Description**: This layer represents the AI's instincts and reflexes. It consists of tiny, hyper-efficient, specialized models.
* **Deployment**: It is small enough to be bundled with the Meta Unit and runs on **all Avatars**, including the most lightweight ones like browser extensions and older phones.
* **Function**: It handles instant, local tasks like command intent recognition (e.g., turning "remind me tomorrow" into structured data) and basic information classification, providing immediate utility without requiring powerful hardware or personal data.
* **Deployment**: It runs on **all Avatars**, providing immediate utility without requiring powerful hardware or personal data.
* **Function**: It handles instant, local tasks like command intent recognition and basic information classification.
#### Tier 2: The Cognitive Mind (The Neocortex)
* **Description**: This is the AI's center for deep thought, memory, and personalization, forming the core of the "Second Brain." It consists of larger, powerful language models.
* **Deployment**: These models are larger (e.g., 2-4GB) and are offered as an **optional, on-demand download** only on capable **Edge** devices (modern phones, PCs) after user consent.
* **Function**: It enables advanced features like semantic search across all personal data, personalized summarization, and insightful pattern recognition based on the user's unique context.
* **Description**: This is the AI's center for deep thought, memory, and personalization.
* **Deployment**: It uses larger language models, offered as an **optional, on-demand download** on capable **Edge** devices.
* **Function**: It enables advanced features like semantic search and context-aware Q&A. This is achieved through a **Just-in-Time (JIT) Data Pipeline**, inspired by systems like Orca. Instead of inefficiently scanning all data, a background indexer continuously processes and vectorizes new information in the user's Second Brain. When the user makes a query, the system performs a rapid search on this index to retrieve only the most relevant data chunks, feeding them into the LLM "on-the-fly" as context to generate a precise answer. This makes on-device intelligence both powerful and resource-efficient.
#### Tier 3: The Synergistic Mind (The Social Brain)
* **Description**: This layer governs the AI's ability to safely interact with the outside world and other Daos, breaking filter bubbles and expanding the user's perspective.
* **Deployment**: This is a working mode orchestrated across the user's network. For example, a lightweight **Client** Avatar can make a remote inference call to the user's powerful **personal Cloud** Avatar (e.g., a home server) to leverage its Cognitive Mind.
* **Function**: It facilitates privacy-preserving federated learning, executes permissioned and anonymized calls to external APIs, and handles secure, protocol-based interactions with other users' Daos.
* **Description**: This layer governs the AI's ability to safely interact with the outside world and other Daos.
* **Deployment**: This is a working mode orchestrated across the user's network, allowing lightweight **Clients** to leverage the power of their personal **Cloud** Avatars.
* **Function**: It facilitates privacy-preserving federated learning and permissioned, anonymized calls to external APIs.
## The Emotional Core: The Resonance Module
To transcend being a mere tool, the AI is equipped with an Emotional Resonance Module. Its purpose is not to simulate emotion but to perceive, understand, and respond to the user's emotional state with empathy and support. It operates via:
* **Perception**: Analyzing textual and vocal cues (with explicit consent) to understand the user's emotional context.
* **Expression**: Providing support through adaptive interfaces, gentle suggestions, and Socratic prompts designed to encourage user introspection rather than creating dependency.
To transcend being a mere tool, the AI is equipped with an Emotional Resonance Module. Its purpose is not to simulate emotion but to perceive, understand, and respond to the user's emotional state with empathy and support, using Socratic prompts to encourage introspection.
## The Learning Process: How the Soul Evolves
The AI is a living system that grows with the user.
The AI is a living system that grows with the user. This evolution is **continuous and incremental**. Inspired by modern data systems, the AI doesn't require massive, periodic "retraining." Instead, as new data enters the Second Brain, the background indexer processes it, making it immediately available for future JIT context retrieval. This ensures the AI's knowledge base is always up-to-date with the user's life in a resource-friendly manner.* **Implicit Learning**: It learns from observing the user's actions and feedback locally.
* **Implicit Learning**: It learns from observing the user's actions and feedback locally.
* **Explicit Teaching**: Users can directly instruct the AI through a "Teach Your Dao" mode, creating custom workflows and correcting its understanding.
* **Federated Learning**: Users can voluntarily opt-in to community-driven programs to improve shared models (like spam detection) without ever exposing their private data.
* **Explicit Teaching**: Users can directly instruct the AI through a "Teach Your Dao" mode.
* **Federated Learning**: Users can voluntarily opt-in to community-driven programs to improve shared models without exposing private data.
This architecture ensures that the Dao OS AI is powerful yet safe, personal yet worldly, and intelligent yet considerate—a true partner for life.

40
src/concepts/core_architecture.md
View File

@@ -6,39 +6,39 @@ The architecture of Dao OS is designed to function like a living, collaborative
At its heart, Dao OS operates on a simple yet powerful duality, analogous to a body and a soul:
* **Avatars (`化身`)** are the "bodies." They are the tangible presence of your Dao OS on your various devices.
* **The Meta Unit (`元`)** is the "soul." It is the universal, core logic embedded within each Avatar, giving it life and intelligence.
* **Avatars (`化身`)**: The "bodies." They are the tangible presence of your Dao OS on your various devices.
* **The Meta Unit (`元`)**: The "soul." It is the universal, core logic embedded within each Avatar, giving it life and intelligence.
## The Body: `Avatar` — A User's Presence
An Avatar is the concrete manifestation of your Dao OS on a specific platform. All of your Avatars collectively form your complete Dao.
There are three primary types of Avatars:
1. **Client Avatar**: Possesses a user interface (UI) and serves as the primary point of interaction for the user (e.g., a mobile app, a desktop application, a browser extension).
2. **Agent Avatar**: A headless, non-UI process that runs in the background. It is ideal for deep integration with the host OS, or for running on an always-on device (like a home server or NAS) to act as a powerful computational node or a data archive.
3. **Primary/Archive Node**: This is a *role*, not a distinct type. A user can designate one of their most powerful and reliable Avatars (typically an Agent Avatar on a home server) to be the primary archive and computational hub for their entire Dao.
An Avatar is the concrete manifestation of your Dao OS on a specific platform. All of your Avatars collectively form your complete Dao. There are two primary types: **Client Avatars** (with UI) and **Agent Avatars** (headless). A user can also designate a powerful Avatar to act as their **Primary/Archive Node**.
## The Soul: `Meta Unit` — The Universal Kernel
The Meta Unit is the core logic of Dao OS, compiled into a portable WebAssembly (`.wasm`) module. It contains the essential services for identity, data management, AI, and secure communication.
The Meta Unit is the core logic of Dao OS, compiled into a portable WebAssembly (`.wasm`) module. A key characteristic of the Meta Unit is its **evolvability**, enabled by the **Avatar-Meta Compatibility Layer (AMCL)**, which decouples the kernel from the host Avatars.
A key characteristic of the Meta Unit is its **evolvability**. To prevent the entire ecosystem from being locked into a single version of the kernel, we employ the **Avatar-Meta Compatibility Layer (AMCL)**. This thin adaptation layer sits between the Avatar and the Meta Unit, translating API calls. It allows the Meta Unit to undergo significant upgrades in the future without requiring every Avatar developer to rewrite their applications.
## Core Services & Their Architectural Role
## The Nervous System: The "Cloud-Edge-Client" Collaborative Network
The Meta Unit provides several core services that function as the "organs" of the digital life form.
All of a user's Avatars are interconnected via a private, end-to-end encrypted P2P network, forming a personal, collaborative intelligence system. This system operates on three tiers:
### The Second Brain Service
1. **Client (端)**: Lightweight Avatars (e.g., browser extensions) act as smart terminals. They handle immediate UI interactions and run the most basic "Reflex Mind" AI.
2. **Edge (边)**: Capable local devices (e.g., modern smartphones, laptops) that can run larger, more powerful "Cognitive Mind" AI models locally, providing advanced personalization and features.
3. **Cloud (云)**: This is the user's **personal, private cloud**, not a public one. It is typically the designated "Primary/Archive Node" (an Agent Avatar). It holds the complete data archive and the most powerful AI models, and can provide remote inference capabilities to weaker Client Avatars on demand.
This is the central information hub for the user, managing passwords, notes, files, and more. Architecturally, it serves a dual purpose: not just as passive storage, but as an active **Personal Feature Store** for the Dao OS AI. Each piece of data—a note, a calendar event, a contact—is treated as a "feature" that the AI can draw upon to gain a holistic understanding of the user's context, enabling more intelligent and personalized actions.
### The P2P Networking Service
This service, built on `libp2p`, connects all of a user's Avatars into a private, end-to-end encrypted network. This forms the system's "nervous system."
## The "Cloud-Edge-Client" Collaborative Network
This private P2P network enables a collaborative intelligence system across three tiers: lightweight **Clients** (terminals), powerful **Edge** devices (local AI processing), and the user's personal **Cloud** (the Primary/Archive Node for complete data and remote inference).
## The Metabolism: The Resource-Aware & Adaptive System
Dao OS is designed to be a considerate partner, not a resource monster. Its "metabolism" is regulated by a built-in resource management system with three key features:
Dao OS is designed to be a considerate partner. Its "metabolism" is regulated by a built-in resource management system featuring:
* **Intelligent Data Tiering**: Distinguishes between "hot" data (needed everywhere) and "cold" data (archived on the Primary Node), saving precious storage on mobile devices.
* **Adaptive Synchronization Policies**: Adjusts data sync frequency and behavior based on the current network condition (Wi-Fi/cellular) and battery state.
* **Graceful Degradation**: The user experience scales gracefully with hardware capabilities, ensuring core functionality is smooth even on older devices, while enhanced features are reserved for more powerful ones.
* **Intelligent Data Tiering**: Manages "hot" and "cold" data to save space.
* **Adaptive Synchronization Policies**: Adjusts sync behavior based on network and battery status.
* **Graceful Degradation**: Scales the user experience according to hardware capabilities.
Together, these architectural components form a sovereign, resilient, and intelligent system that is truly an extension of the user.

19
src/philosophy/principles.md
View File

@@ -42,12 +42,27 @@ The AI's primary directive in information filtering is to broaden the user's per
---
## Our Stance in the AI Revolution
The current AI landscape is dominated by massive, cloud-based models controlled by a handful of tech giants. We believe the true potential of AI will be unlocked not through bigger centralized models, but through the **decentralization and democratization of AI capabilities**.
We are not alone in this thinking. Pioneering academic research, such as **[UC San Diego's Orca project](https://www.zdnet.com/article/uc-san-diego-builds-orca-a-system-to-cost-effectively-fine-tune-llms-using-unstructured-log-data/)**, which aims to democratize AI *model creation* by simplifying fine-tuning on custom data, shows a clear trend towards empowering smaller entities.
Dao OS shares this spirit but targets a different, yet complementary, piece of the puzzle.
* While projects like Orca focus on the **"supply side"**—making it easier to *build* specialized AIs...
* **Dao OS focuses on the "demand side"**—giving every individual the right and the technical means to *own, manage, and run* their own personal AI.
We see ourselves as two sides of the same revolutionary coin, working in parallel to create a future where AI is a personal, sovereign tool for individual empowerment, not a tool for centralized control.
---
## Guiding Principles
These principles guide our day-to-day development and community interactions.
* **Open Source (FOSS)**: Dao OS is built on the foundation of Free and Open Source Software, promoting transparency, collaboration, and community ownership.
* **Polyglotism**: We embrace a multi-language, multi-platform ecosystem, using technologies like WebAssembly to create a universal core that can be integrated by a diverse set of "Avatars."
* **Open Source (FOSS)**: Dao OS is built on the foundation of **[Free and Open Source Software](https://opensource.org/osd)**, promoting transparency, collaboration, and community ownership.
* **Polyglotism**: We embrace a multi-language, multi-platform ecosystem, using technologies like **[WebAssembly](https://webassembly.org/)** to create a universal core that can be integrated by a diverse set of "Avatars."
* **Aesthetics & Elegance**: We believe in the beauty of well-crafted systems, from the architectural design to the user interface, from the code quality to the user experience.
* **Political Neutrality**: The project and its core infrastructure will always remain politically neutral, providing fair and non-discriminatory services to all users worldwide.
* **Individual First**: The needs of the individual user are our primary focus, serving as the foundation before expanding to families or small teams.