opc-agent 4.0.0 → 4.0.2

package/README.md

@@ -2,19 +2,19 @@
 
 # ⚡ OPC Agent
 
-### 内置记忆进化的 TypeScript Agent 框架
+### The Self-Evolving Agent Runtime — 自进化 Agent 运行时
 
-[![npm version](https://img.shields.io/badge/npm-v3.0.0-blue)](https://www.npmjs.com/package/opc-agent)
+[![npm version](https://img.shields.io/badge/npm-v4.0.0-blue)](https://www.npmjs.com/package/opc-agent)
 [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://github.com/Deepleaper/opc-agent/blob/main/LICENSE)
-[![Tests](https://img.shields.io/badge/tests-204_passing-brightgreen)]()
+[![Tests](https://img.shields.io/badge/tests-1024%2B_passing-brightgreen)]()
 [![TypeScript](https://img.shields.io/badge/TypeScript-5.7-3178C6?logo=typescript&logoColor=white)](https://www.typescriptlang.org/)
 [![Node.js](https://img.shields.io/badge/Node.js-%3E%3D18-339933?logo=node.js&logoColor=white)](https://nodejs.org/)
 
-**你的 Agent 不该只是"能跑"——它应该能记住、能进化、能跨 11 个渠道触达用户。**
+**The Self-Evolving Agent Stack 的运行引擎 — 让每个 Agent 都能学习、记忆、进化。**
 
-OPC Agent 是一个全生命周期 Agent 框架：从 `opc init` 创建，到 `opc chat` 对话，到 `opc studio` 可视化管理，<br>内置记忆进化引擎让 Agent **越用越聪明**，而不是每次从零开始。
+OPC Agent 是自进化 Agent 技术栈中的 **Agent Runtime**：从 `opc init` 创建，到 `opc chat` 对话，到 `opc studio` 可视化管理。<br>内置 learn → recall → evolve 闭环，让 Agent **越用越聪明**——不是每次从零开始，而是持续积累和进化。<br>开发者使用路径：① 选模板 (Agent Templates) → ② 接模型 (Agent Model Layer) → ③ **跑起来 (Agent Runtime)** → ④ 自动进化 (Agent Memory)。
 
-[快速开始](#-快速开始) · [核心特性](#-核心特性) · [代码示例](#-代码示例) · [CLI 参考](#-cli-参考) · [English](#-english-version)
+[快速开始](#-快速开始) · [🧬 自进化](#-自进化-self-evolution) · [核心特性](#-核心特性) · [代码示例](#-代码示例) · [CLI 参考](#-cli-参考) · [English](#-english-version)
 
 </div>
 
@@ -22,6 +22,22 @@ OPC Agent 是一个全生命周期 Agent 框架：从 `opc init` 创建，到 `o
 
 ## 🚀 快速开始
 
+### ⚡ 一键安装 / One-Line Install
+
+**macOS / Linux:**
+```bash
+curl -fsSL https://raw.githubusercontent.com/Deepleaper/opc-agent/main/install.sh | bash
+```
+
+**Windows (PowerShell):**
+```powershell
+irm https://raw.githubusercontent.com/Deepleaper/opc-agent/main/install.ps1 | iex
+```
+
+> 脚本会自动安装 Node.js、OPC Agent，可选安装 Ollama 本地模型。支持 `--yes` 跳过确认，`--no-ollama` 跳过 Ollama。
+
+### 手动安装 / Manual Install
+
 ```bash
 npm install -g opc-agent
 opc init my-agent --role customer-service
@@ -48,18 +64,82 @@ Agent: 您上次查询的订单 #12300 已签收，签收时间是 4 月 15 日
 
 | | 特性 | 说明 |
 |---|---|---|
-| 🧠 | **记忆进化** | learn → recall → evolve，内置知识自动沉淀 |
+| 🧬 | **自进化 (Self-Evolution)** | learn → recall → evolve 闭环，Agent 越用越聪明，知识自动积累和提炼 |
+| 🧠 | **记忆进化** | 内置 DeepBrain，知识自动沉淀、聚类、去重、精炼 |
 | 🔧 | **20+ CLI 命令** | init / chat / run / start / studio / doctor / eval / traces / publish |
-| 📡 | **11 种 Channel** | Telegram / Discord / Slack / WeChat / Feishu / Email / Web / WebSocket / Voice / Webhook / API |
+| 📡 | **25 种 Channel** | Telegram / Discord / Slack / WeChat / Feishu / Email / Web / WebSocket / Voice / Webhook / API / WhatsApp / LINE / Teams / SMS / DingTalk / Twitter / Instagram / Facebook / Signal / Matrix / XMPP / IRC / Twilio / Custom |
 | 🔌 | **三大协议** | Google A2A + AG-UI + MCP（Server & Client） |
 | 🎨 | **OPC Studio** | 可视化管理后台，一条命令 `opc studio` 启动 |
 | 📊 | **OpenTelemetry** | 全链路追踪 + p50 / p95 / p99 延迟指标 |
 | 🧪 | **内置评估** | `opc eval` 运行 24 个评估用例，量化 Agent 质量 |
 | 🔍 | **RAG Pipeline** | 5 种分块策略 + 4 种重排序（通过 DeepBrain） |
 | 📦 | **打包分发** | `opc publish` 一键发布到 npm |
-| 🏭 | **工位模板** | 100+ 专业角色，`opc init --role` 秒级创建 |
-| 🔒 | **安全沙箱** | 命令审批 + API Key 加密 + 文件/网络限制 |
+| 🏭 | **角色模板** | 100+ 专业角色，`opc init --role` 秒级创建 |
+| 🔒 | **安全全家桶** | 命令审批 + API Key 加密 + 文件/网络限制 + 沙箱隔离 |
 | 🤖 | **子 Agent** | spawn / parallel / kill 多 Agent 协作 |
+| 🌐 | **浏览器自动化** | Playwright 集成，网页交互、截图、数据抓取 |
+| 👁️ | **Vision** | 多模态图像理解，截图分析、文档识别 |
+| 🎙️ | **语音通话** | 实时语音对话，支持 TTS + STT |
+| 🏠 | **Home Assistant** | 智能家居集成，控制 IoT 设备 |
+| 💻 | **IDE Bridge** | VS Code / Cursor 深度集成 |
+| 🔗 | **Node Network** | 多节点网络，Agent 跨设备协作 |
+| 🚪 | **Gateway** | 统一网关，安全路由 + 负载均衡 |
+| 🗜️ | **Context 压缩** | 智能上下文压缩，长对话不丢关键信息 |
+| 🌐 | **API Server** | REST API 服务，外部系统集成 |
+| 📎 | **@引用** | @mention 引用其他 Agent 或资源 |
+| 📋 | **Session Manager** | 会话管理，多会话并行、历史回溯 |
+| 🎯 | **RL（强化学习）** | 基于反馈的策略优化 |
+
+---
+
+## 🧬 自进化 (Self-Evolution)
+
+**OPC 的核心卖点——自进化，这是跟所有其他 Agent 框架的根本区别。**
+
+别人的 Agent 是静态的——用完就忘，换个人接手一切归零。我们的 Agent 会学习、会记忆、会进化。
+
+### 1. 你搭建的每个 Agent 都会自进化
+
+通过 learn → recall → evolve 闭环，Agent 自动积累知识，越用越聪明：
+
+```
+┌──────────────────────────────────────────────────────┐
+│                                                      │
+│   📥 learn        🔍 recall        🧬 evolve         │
+│   每次交互后       需要时自动        定期自动提炼       │
+│   存储经验         检索相关记忆      知识越来越精        │
+│                                                      │
+│   Day 1:  "这个问题我没见过"                           │
+│   Day 7:  "类似问题上次这样解决的"                      │
+│   Day 30: "这类问题的最佳方案是..."                     │
+│                                                      │
+└──────────────────────────────────────────────────────┘
+```
+
+**具体例子**：一个客服 Agent——
+- **第 1 天**：只能照着 prompt 回答，遇到没见过的问题需要人工介入
+- **第 7 天**：自动积累了 200+ 条对话经验，`evolve()` 提炼出 15 条产品知识
+- **第 30 天**：知识库精炼为 5 条核心洞察 + 50 条常见问题最佳回答，准确率从 60% 提升到 95%
+
+```typescript
+import { AgentRuntime } from 'opc-agent';
+
+const runtime = new AgentRuntime('./agent.yaml');
+await runtime.start();
+
+// 不需要你做任何事——Agent 每次对话自动：
+// 1. recall() — 检索相关历史记忆
+// 2. respond() — 基于记忆生成更好的回答
+// 3. learn()  — 把这次交互存入记忆
+// 4. evolve() — 定期把零散经验提炼为结构化知识
+```
+
+### 2. 这个开源项目本身也在自进化
+
+OPC 不是一个静态的框架——它是一个**活的生态**：
+- 🌱 **社区驱动**：每一个 PR、Issue、讨论都在推动四件套进化
+- 🔄 **持续迭代**：每天自动集成最新的 AI 技术和最佳实践
+- 📈 **知识飞轮**：越多 Agent 使用 → 越多经验沉淀 → 框架越来越强 → 新 Agent 起步越高
 
 ---
 
@@ -73,7 +153,7 @@ Agent: 您上次查询的订单 #12300 已签收，签收时间是 4 月 15 日
 │ DeepBrain  │ AgentKits  │Workstation │ OPC Core  │
 │ 🧠 记忆进化 │ 📊 统一模型 │ 👤 角色模板 │ ⚡ 运行引擎 │
 │            │            │            │           │
-│ learn()    │ OpenAI     │ 100+ 角色  │ 11 渠道    │
+│ learn()    │ OpenAI     │ 100+ 角色  │ 25 渠道    │
 │ recall()   │ Anthropic  │ YAML 定义  │ 3 大协议   │
 │ evolve()   │ Ollama     │ 技能系统   │ Cron 调度  │
 │ RAG        │ DeepSeek   │ 一键创建   │ 子 Agent   │
@@ -154,9 +234,39 @@ memory:
 
 ---
 
+## 🔄 双闭环知识系统 (Dual-Loop Knowledge)
+
+OPC Agent 的知识系统由两个闭环组成：
+
+```
+┌─────────────────────────────────────────────────────┐
+│  小闭环（本地，免费）                                  │
+│  Agent 本地 learn → recall → evolve                  │
+│  离线也能用，数据完全在你手里                           │
+│                                                     │
+│  大闭环（Hub，增值）                                   │
+│  Agent ↔ Workstation Hub 知识共享                    │
+│  集体智慧 > 个体经验，新 Agent 站在前人肩膀上            │
+└─────────────────────────────────────────────────────┘
+```
+
+**本地是主人，Hub 是助手**——没有网络也能用，联网后自动同步和进化。
+
+```
+agentkits (Model Layer) — 调 LLM
+     ↕
+opc-agent (Runtime) — 跑 Agent（本地）  ← 你在这里
+     ↕
+deepbrain (Memory Engine) — 存知识（引擎）
+     ↕
+agent-workstation (Knowledge Platform) — 知识生命周期（Hub）
+```
+
+---
+
 ## 🌱 Brain Seed 自动加载
 
-v2.1.0 起，OPC Agent 支持三层知识种子自动加载。当 `brain-seeds/` 目录存在时，Agent 首次启动自动导入行业→岗位→工位知识：
+v2.1.0 起，OPC Agent 支持三层知识种子自动加载。当 `brain-seeds/` 目录存在时，Agent 首次启动自动导入行业→岗位→模板知识：
 
 ```typescript
 const agent = new BaseAgent({
@@ -174,34 +284,79 @@ agent.setLongTermMemory(brain);
 ```bash
 opc brain seed              # 查看当前 brain seed 状态
 opc brain seed --load       # 手动加载/重新加载种子
-opc brain evolve            # 触发知识进化（工位→岗位→行业）
+opc brain evolve            # 触发知识进化（template → job → industry）
 opc brain status            # 查看 brain 统计信息
 ```
 
 ---
 
-## 📊 对比
-
-| 特性 | OPC Agent | Mastra | CrewAI | LangGraph |
-|------|:---------:|:------:|:------:|:---------:|
-| 语言 | TypeScript | TypeScript | Python | Python |
-| 记忆进化 | ✅ 内置 | ❌ | ⚠️ 有记忆模块 | ⚠️ 可通过 checkpointer 实现 |
-| CLI 全生命周期 | ✅ 20+ 命令 | ⚠️ 部分 | ⚠️ CLI 有限 | ❌ |
-| 渠道集成 | 11 内置 | ⚠️ 需自行接入 | ⚠️ 需自行接入 | ⚠️ 需自行接入 |
-| A2A 协议 | ✅ | ⚠️ | ❌ | ❌ |
-| AG-UI 协议 | ✅ | ❌ | ❌ | ❌ |
-| MCP (Server+Client) | ✅ | ⚠️ Client | ⚠️ 有集成 | ⚠️ Client |
-| 可视化管理 | ✅ OPC Studio | ⚠️ 有 Dashboard | ⚠️ 有 Dashboard | ✅ LangGraph Studio |
-| 内置评估 | ✅ `opc eval` | ❌ | ❌ | ⚠️ LangSmith 集成 |
-| OpenTelemetry | ✅ | ⚠️ 有 Observability | ❌ | ⚠️ LangSmith |
-| 多 Agent 协作 | ✅ spawn/parallel | ⚠️ 有 Agent Network | ✅ Crew 编排 | ✅ 图编排 |
-| YAML 声明式 | ✅ | ⚠️ | ❌ | ❌ |
-| 社区与生态 | 早期项目 | ✅ 活跃社区 | ✅ 大型社区 | ✅ LangChain 大生态 |
-| 文档完善度 | 🚧 持续完善中 | ✅ | ✅ | ✅ |
-
-OPC Agent 的特色是**记忆进化 + 全生命周期 CLI + 多渠道开箱即用**。各框架各有侧重，建议根据技术栈和场景选择。
-
-> 对比基于各项目公开文档（截至 2026 年 4 月），如有偏差欢迎 [Issue 指正](https://github.com/Deepleaper/opc-agent/issues)。
+## 🔍 与其他 Agent 框架的差异
+
+> **一句话：别人的 Agent 是静态的，OPC 的 Agent 会自进化。**
+
+OPC Agent 的核心设计哲学是**自进化 + 全渠道 + 角色模板 + 生产工具链**。作为自进化 Agent 技术栈（The Self-Evolving Agent Stack）中的 **Agent Runtime**，OPC 更关注的是：一个 Agent 如何在真实生产环境中**持续自进化、随处触达、开箱即用**。
+
+**🔄 自进化飞轮**：Templates 自带 Brain Seed → Model Layer 每次调用自动 learn → Runtime 持续运行 → Memory 自动 evolve → Agent 越来越强。
+
+#### vs Hermes Agent
+
+**设计思路差异**：Hermes 是 Python 单体框架，围绕 Nous Research 的模型生态构建，强调自学习进化和 RL 训练。OPC 是 TypeScript 模块化四件套（Brain / Agent / Channel / CLI），强调工程化和开箱即用。
+
+**OPC 的优势**：记忆进化三件套（learn → recall → evolve 自动聚类提炼）、三层 Brain Seed 预置知识体系、25 个内置渠道（vs Hermes 16+）、100+ 角色模板、原生 A2A + AG-UI 协议支持、OpenAI 兼容 API Server。
+
+**Hermes 的优势**：Python 生态更成熟、GRPO + LoRA 的 RL 训练管线更完善、Nous Research 社区生态更大、Docker 沙箱隔离方案更成熟。
+
+**适合谁**：如果你需要 Python 生态和 RL 训练能力，选 Hermes；如果你需要 TypeScript 全栈、记忆进化、全渠道触达和生产工具链，选 OPC。
+
+#### vs OpenClaw
+
+**设计思路差异**：OpenClaw 是重量级 runtime + 配置驱动，提供完整的 Gateway 运行时环境，适合"安装一次、配置运行"的场景。OPC 是轻量 CLI + 代码优先，四件套可独立使用，适合开发者自由组合。
+
+**OPC 的优势**：模块化四件套可按需引入、100+ 角色模板、记忆进化系统、25 个内置渠道、`opc init/dev/test/deploy` 全生命周期 CLI、A2A/AG-UI 协议支持。
+
+**OpenClaw 的优势**：开箱即用的 runtime 体验、Puppeteer 浏览器自动化、成熟的权限和安全沙箱、配置驱动上手门槛低。
+
+**适合谁**：如果你偏好"配置即运行"的一体化方案，选 OpenClaw；如果你偏好代码优先、模块自由组合、需要记忆进化和模板体系，选 OPC。
+
+#### vs CrewAI
+
+**设计思路差异**：CrewAI 采用 Crew 编排模式——定义 Agent 角色、分配 Task、按流程执行。OPC 采用模板（Template）+ 记忆（Memory）模式——每个 Agent 有独立记忆和角色模板，通过 DeepBrain 实现知识积累和进化。
+
+**OPC 的优势**：记忆进化（不只是短期/长期记忆，而是自动聚类提炼）、Brain Seed 预置知识、25 个内置渠道（CrewAI 需自行接入）、100+ 角色模板、TypeScript 原生。
+
+**CrewAI 的优势**：Crew + Flow 编排模式直观易懂、Python 生态、100K+ 用户的成熟社区、Enterprise 版本功能丰富。
+
+**适合谁**：如果你需要多 Agent 任务编排且偏好 Python，选 CrewAI；如果你需要单 Agent 深度进化、全渠道部署、TypeScript 生态，选 OPC。
+
+#### vs AutoGen
+
+**设计思路差异**：AutoGen 是微软主导的对话驱动多 Agent 框架，核心是 Agent 之间的对话协作和 UserProxy 人机交互。OPC 是任务驱动的模板模式，强调单 Agent 的记忆进化和生产部署能力。
+
+**OPC 的优势**：记忆进化系统、Brain Seed、25 个内置渠道、100+ 角色模板、全生命周期 CLI、YAML 声明式配置。
+
+**AutoGen 的优势**：微软生态支持、分布式 Agent 能力、AutoGen Studio 可视化、UserProxy 人机交互模式成熟、Python/C# 双语言支持、社区活跃。
+
+**适合谁**：如果你需要多 Agent 对话协作和微软生态集成，选 AutoGen；如果你需要记忆进化、全渠道触达、角色模板和 TypeScript 工具链，选 OPC。
+
+#### 功能速查表
+
+| 功能 | OPC Agent | Hermes Agent | OpenClaw | CrewAI | AutoGen |
+|---|:-:|:-:|:-:|:-:|:-:|
+| **语言** | TypeScript | Python | TypeScript | Python | Python/C# |
+| **自进化 (Self-Evolution)** | ✅ learn→recall→evolve | 🔶 | ❌ | ❌ | ❌ |
+| **Brain Seed 预置知识** | ✅ 三层 | ❌ | ❌ | ❌ | ❌ |
+| **内置渠道** | 25 | 16+ | 2 | ❌ | ❌ |
+| **角色模板** | 100+ | ❌ | ❌ | ❌ | ❌ |
+| **CLI 全生命周期** | ✅ 20+ 命令 | 🔶 | 🔶 | 🔶 | 🔶 |
+| **A2A + AG-UI** | ✅ | 🔶 / ❌ | ❌ | ❌ | ❌ |
+| **多 Agent 协作** | ✅ | ✅ | 🔶 | ✅ | ✅ |
+| **RL 训练** | ✅ 反馈优化 | ✅ GRPO+LoRA | ❌ | ❌ | ❌ |
+| **浏览器自动化** | ✅ | ✅ | ✅ | ❌ | ❌ |
+| **可视化** | ✅ OPC Studio | ✅ Dashboard | ❌ | 🔶 | ✅ Studio |
+| **社区成熟度** | 🚧 早期 | ✅ 活跃 | 🚧 小众 | ✅ 大规模 | ✅ 大规模 |
+| **许可证** | Apache-2.0 | MIT | MIT | Apache-2.0 | MIT |
+
+> 对比基于各项目公开文档（截至 2026 年 4 月）。各框架都在快速迭代，如有偏差欢迎 [Issue 指正](https://github.com/Deepleaper/opc-agent/issues)。
 
 ---
 
@@ -209,10 +364,10 @@ OPC Agent 的特色是**记忆进化 + 全生命周期 CLI + 多渠道开箱即
 
 | 包 | 功能 | 安装 |
 |---|---|---|
-| **[opc-agent](https://www.npmjs.com/package/opc-agent)** | Agent OS — 创建、运行、管理 | `npm i opc-agent` |
-| **[deepbrain](https://www.npmjs.com/package/deepbrain)** | 组织大脑 — 记忆存储与进化 | `npm i deepbrain` |
-| **[agentkits](https://www.npmjs.com/package/agentkits)** | 模型层 — 统一 API + 推荐 | `npm i agentkits` |
-| **[agent-workstation](https://www.npmjs.com/package/agent-workstation)** | 工位模板 — 100+ 专业角色 | `npm i agent-workstation` |
+| **[opc-agent](https://www.npmjs.com/package/opc-agent)** | Agent Runtime — 创建、运行、管理 | `npm i opc-agent` |
+| **[deepbrain](https://www.npmjs.com/package/deepbrain)** | Agent Memory — 自进化记忆引擎 | `npm i deepbrain` |
+| **[agentkits](https://www.npmjs.com/package/agentkits)** | Agent Model Layer — 一行代码接入 LLM + 自带记忆 | `npm i agentkits` |
+| **[agent-workstation](https://www.npmjs.com/package/agent-workstation)** | Agent Templates — 100+ 专业角色模板 | `npm i agent-workstation` |
 
 ---
 
@@ -240,7 +395,7 @@ OPC Agent 的特色是**记忆进化 + 全生命周期 CLI + 多渠道开箱即
 | `opc analytics` | 数据分析面板 |
 | `opc brain` | 查看记忆状态 |
 | `opc brain seed` | 查看/加载三层知识种子 |
-| `opc brain evolve` | 触发知识进化（工位→岗位→行业） |
+| `opc brain evolve` | 触发知识进化（template → job → industry） |
 | `opc brain status` | 查看 brain 详细统计 |
 | `opc jobs` | 查看定时任务 |
 | `opc skills` | 查看已学技能 |
@@ -264,7 +419,7 @@ OPC Agent 的特色是**记忆进化 + 全生命周期 CLI + 多渠道开箱即
 
 ---
 
-## 📡 11 种 Channel
+## 📡 25 种 Channel
 
 | 渠道 | 状态 | 说明 |
 |------|:----:|------|
@@ -275,10 +430,24 @@ OPC Agent 的特色是**记忆进化 + 全生命周期 CLI + 多渠道开箱即
 | 📧 Email | ✅ | IMAP / SMTP |
 | 💚 WeChat | ✅ | 企业微信 / 个人微信 |
 | 🔵 Feishu | ✅ | 飞书机器人 |
-| 🎤 Voice | ✅ | 语音通话 |
+| 🎤 Voice | ✅ | 语音通话（TTS + STT） |
 | 🔌 WebSocket | ✅ | 实时双向通信 |
 | 🪝 Webhook | ✅ | HTTP 回调 |
 | 📡 REST API | ✅ | HTTP API |
+| 💬 WhatsApp | ✅ | WhatsApp Business API |
+| 🟢 LINE | ✅ | LINE Messaging API |
+| 🟣 Teams | ✅ | Microsoft Teams Bot |
+| 📲 SMS | ✅ | Twilio / 云通信 |
+| 🔷 DingTalk | ✅ | 钉钉机器人 |
+| 🐦 Twitter/X | ✅ | DM + Mentions |
+| 📸 Instagram | ✅ | Instagram DM |
+| 📘 Facebook | ✅ | Messenger API |
+| 🔐 Signal | ✅ | Signal Bot |
+| 🟩 Matrix | ✅ | Matrix 协议 |
+| 💬 XMPP | ✅ | Jabber/XMPP |
+| 💻 IRC | ✅ | IRC 协议 |
+| 📞 Twilio | ✅ | 电话 / IVR |
+| 🔧 Custom | ✅ | 自定义渠道适配器 |
 
 ---
 
@@ -310,19 +479,19 @@ npm test
 
 # ⚡ OPC Agent
 
-### TypeScript Agent Framework with Built-in Memory Evolution
+### The Self-Evolving Agent Runtime — Every Agent You Build Learns, Remembers, and Evolves
 
-[![npm version](https://img.shields.io/badge/npm-v3.0.0-blue)](https://www.npmjs.com/package/opc-agent)
+[![npm version](https://img.shields.io/badge/npm-v4.0.0-blue)](https://www.npmjs.com/package/opc-agent)
 [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://github.com/Deepleaper/opc-agent/blob/main/LICENSE)
-[![Tests](https://img.shields.io/badge/tests-204_passing-brightgreen)]()
+[![Tests](https://img.shields.io/badge/tests-1024%2B_passing-brightgreen)]()
 [![TypeScript](https://img.shields.io/badge/TypeScript-5.7-3178C6?logo=typescript&logoColor=white)](https://www.typescriptlang.org/)
 [![Node.js](https://img.shields.io/badge/Node.js-%3E%3D18-339933?logo=node.js&logoColor=white)](https://nodejs.org/)
 
-**Your Agent shouldn't just "run" — it should remember, evolve, and reach users across 11 channels.**
+**The runtime engine of The Self-Evolving Agent Stack — every Agent learns, remembers, and evolves.**
 
-OPC Agent is a full-lifecycle Agent framework: from `opc init` to create, `opc chat` to converse, `opc studio` to manage visually.<br>The built-in memory evolution engine makes your Agent **smarter over time**, instead of starting from scratch every session.
+OPC Agent is the **Agent Runtime** in the self-evolving Agent stack: from `opc init` to create, `opc chat` to converse, `opc studio` to manage visually.<br>The built-in learn → recall → evolve loop makes your Agent **smarter over time** — not starting from scratch, but continuously accumulating and evolving knowledge.<br>Developer path: ① Pick a template (Agent Templates) → ② Connect models (Agent Model Layer) → ③ **Run it (Agent Runtime)** → ④ Auto-evolve (Agent Memory).
 
-[Quick Start](#-quick-start) · [Features](#-features) · [Code Examples](#-code-examples-1) · [CLI Reference](#-cli-reference-1)
+[Quick Start](#-quick-start) · [🧬 Self-Evolution](#-self-evolution) · [Features](#-features) · [Code Examples](#-code-examples-1) · [CLI Reference](#-cli-reference-1)
 
 </div>
 
@@ -356,9 +525,10 @@ Agent: Your previous order #12300 was delivered on April 15th at 2 PM.
 
 | | Feature | Description |
 |---|---|---|
-| 🧠 | **Memory Evolution** | learn → recall → evolve — built-in automatic knowledge distillation |
+| 🧬 | **Self-Evolution** | learn → recall → evolve loop — Agents get smarter over time, knowledge auto-accumulates and refines |
+| 🧠 | **Memory Evolution** | Built-in DeepBrain — auto distillation, clustering, deduplication, refinement |
 | 🔧 | **20+ CLI Commands** | init / chat / run / start / studio / doctor / eval / traces / publish |
-| 📡 | **11 Channels** | Telegram / Discord / Slack / WeChat / Feishu / Email / Web / WebSocket / Voice / Webhook / API |
+| 📡 | **25 Channels** | Telegram / Discord / Slack / WeChat / Feishu / Email / Web / WebSocket / Voice / Webhook / API / WhatsApp / LINE / Teams / SMS / DingTalk / Twitter / Instagram / Facebook / Signal / Matrix / XMPP / IRC / Twilio / Custom |
 | 🔌 | **3 Protocols** | Google A2A + AG-UI + MCP (Server & Client) |
 | 🎨 | **OPC Studio** | Visual management dashboard — one command `opc studio` |
 | 📊 | **OpenTelemetry** | Full distributed tracing + p50 / p95 / p99 latency metrics |
@@ -366,8 +536,73 @@ Agent: Your previous order #12300 was delivered on April 15th at 2 PM.
 | 🔍 | **RAG Pipeline** | 5 chunking strategies + 4 rerankers (via DeepBrain) |
 | 📦 | **Package & Publish** | `opc publish` — one command to publish to npm |
 | 🏭 | **Role Templates** | 100+ professional roles, `opc init --role` for instant creation |
-| 🔒 | **Security Sandbox** | Command approval + API Key encryption + file/network restrictions |
+| 🔒 | **Security Suite** | Command approval + API Key encryption + file/network restrictions + sandboxing |
 | 🤖 | **Sub-Agents** | spawn / parallel / kill for multi-Agent collaboration |
+| 🌐 | **Browser Automation** | Playwright integration — web interaction, screenshots, data scraping |
+| 👁️ | **Vision** | Multimodal image understanding — screenshot analysis, document recognition |
+| 🎙️ | **Voice Call** | Real-time voice conversation with TTS + STT |
+| 🏠 | **Home Assistant** | Smart home integration, IoT device control |
+| 💻 | **IDE Bridge** | Deep integration with VS Code / Cursor |
+| 🔗 | **Node Network** | Multi-node network, cross-device Agent collaboration |
+| 🚪 | **Gateway** | Unified gateway with secure routing + load balancing |
+| 🗜️ | **Context Compression** | Smart context compression — keep key info in long conversations |
+| 🌐 | **API Server** | REST API server for external system integration |
+| 📎 | **@Mention** | @mention other Agents or resources |
+| 📋 | **Session Manager** | Multi-session parallel, history rollback |
+| 🎯 | **RL (Reinforcement Learning)** | Feedback-based strategy optimization |
+
+---
+
+## 🧬 Self-Evolution
+
+**The core differentiator of OPC — self-evolution. This is what fundamentally separates us from every other Agent framework.**
+
+Other frameworks build static Agents — they forget everything after each session, and when someone else takes over, everything resets. OPC Agents learn, remember, and evolve.
+
+### 1. Every Agent You Build Self-Evolves
+
+Through the learn → recall → evolve loop, Agents automatically accumulate knowledge and get smarter:
+
+```
+┌──────────────────────────────────────────────────────┐
+│                                                      │
+│   📥 learn        🔍 recall        🧬 evolve         │
+│   After each      Auto-retrieve    Periodically      │
+│   interaction     relevant         refine knowledge   │
+│                   memories         — sharper, not     │
+│                                    bigger             │
+│                                                      │
+│   Day 1:  "I haven't seen this issue before"         │
+│   Day 7:  "A similar issue was solved this way"      │
+│   Day 30: "The best practice for this type is..."    │
+│                                                      │
+└──────────────────────────────────────────────────────┘
+```
+
+**Concrete example**: A customer service Agent —
+- **Day 1**: Can only respond based on the initial prompt; needs human escalation for unknown issues
+- **Day 7**: Accumulated 200+ conversation experiences; `evolve()` distilled 15 product knowledge entries
+- **Day 30**: Refined into 5 core insights + 50 best-practice answers; accuracy from 60% → 95%
+
+```typescript
+import { AgentRuntime } from 'opc-agent';
+
+const runtime = new AgentRuntime('./agent.yaml');
+await runtime.start();
+
+// You don't need to do anything — the Agent automatically:
+// 1. recall() — retrieves relevant past memories
+// 2. respond() — generates better answers based on memory
+// 3. learn()  — stores this interaction into memory
+// 4. evolve() — periodically distills scattered experiences into structured knowledge
+```
+
+### 2. The Open-Source Project Itself Self-Evolves
+
+OPC isn't a static framework — it's a **living ecosystem**:
+- 🌱 **Community-driven**: Every PR, Issue, and discussion pushes the suite forward
+- 🔄 **Continuous iteration**: Daily integration of the latest AI techniques and best practices
+- 📈 **Knowledge flywheel**: More Agents → richer experience → stronger framework → new Agents start smarter
 
 ---
 
@@ -381,7 +616,7 @@ Agent: Your previous order #12300 was delivered on April 15th at 2 PM.
 │ DeepBrain  │ AgentKits  │Workstation │ OPC Core  │
 │ 🧠 Memory  │ 📊 Models  │ 👤 Roles   │ ⚡ Engine  │
 │            │            │            │           │
-│ learn()    │ OpenAI     │ 100+ roles │ 11 chan.  │
+│ learn()    │ OpenAI     │ 100+ roles │ 25 chan.  │
 │ recall()   │ Anthropic  │ YAML def.  │ 3 proto.  │
 │ evolve()   │ Ollama     │ Skill sys. │ Cron sched│
 │ RAG        │ DeepSeek   │ One-click  │ Sub-Agent │
@@ -462,6 +697,36 @@ memory:
 
 ---
 
+## 🔄 Dual-Loop Knowledge System
+
+OPC Agent's knowledge system consists of two loops:
+
+```
+┌─────────────────────────────────────────────────────┐
+│  Small Loop (Local, Free)                           │
+│  Agent local learn → recall → evolve                │
+│  Works offline, data stays on your machine          │
+│                                                     │
+│  Big Loop (Hub, Value-Add)                          │
+│  Agent ↔ Workstation Hub knowledge sharing          │
+│  Collective wisdom > individual experience          │
+└─────────────────────────────────────────────────────┘
+```
+
+**Local is the owner, Hub is the helper** — works without internet, auto-syncs when connected.
+
+```
+agentkits (Model Layer) — LLM calls
+     ↕
+opc-agent (Runtime) — run Agents (local)  ← You are here
+     ↕
+deepbrain (Memory Engine) — store knowledge (engine)
+     ↕
+agent-workstation (Knowledge Platform) — knowledge lifecycle (Hub)
+```
+
+---
+
 ## 🌱 Brain Seed Auto-Loading
 
 Since v2.1.0, OPC Agent supports automatic 3-tier knowledge seed loading. When a `brain-seeds/` directory exists, the Agent auto-imports industry → job → workstation knowledge on first run:
@@ -488,28 +753,73 @@ opc brain status            # View brain statistics
 
 ---
 
-## 📊 Comparison
-
-| Feature | OPC Agent | Mastra | CrewAI | LangGraph |
-|---------|:---------:|:------:|:------:|:---------:|
-| Language | TypeScript | TypeScript | Python | Python |
-| Memory Evolution | ✅ Built-in | ❌ | ⚠️ Has memory module | ⚠️ Via checkpointer |
-| Full-lifecycle CLI | ✅ 20+ cmds | ⚠️ Partial | ⚠️ Limited CLI | ❌ |
-| Channel integrations | 11 built-in | ⚠️ DIY | ⚠️ DIY | ⚠️ DIY |
-| A2A Protocol | ✅ | ⚠️ | ❌ | ❌ |
-| AG-UI Protocol | ✅ | ❌ | ❌ | ❌ |
-| MCP (Server+Client) | ✅ | ⚠️ Client | ⚠️ Has integration | ⚠️ Client |
-| Visual Studio | ✅ OPC Studio | ⚠️ Dashboard | ⚠️ Dashboard | ✅ LangGraph Studio |
-| Built-in Eval | ✅ `opc eval` | ❌ | ❌ | ⚠️ LangSmith integration |
-| OpenTelemetry | ✅ | ⚠️ Observability | ❌ | ⚠️ LangSmith |
-| Multi-Agent | ✅ spawn/parallel | ⚠️ Agent Network | ✅ Crew orchestration | ✅ Graph orchestration |
-| YAML Declarative | ✅ | ⚠️ | ❌ | ❌ |
-| Community & ecosystem | Early-stage project | ✅ Active community | ✅ Large community | ✅ LangChain ecosystem |
-| Documentation | 🚧 Improving | ✅ | ✅ | ✅ |
-
-OPC Agent's differentiator is **Memory Evolution + Full-lifecycle CLI + multi-channel out of the box**. Each framework has different strengths — choose based on your tech stack and use case.
-
-> Comparison based on each project's public documentation as of April 2026. Corrections welcome via [Issues](https://github.com/Deepleaper/opc-agent/issues).
+## 🔍 How OPC Agent Differs from Other Frameworks
+
+> **In one sentence: other Agents are static; OPC Agents self-evolve.**
+
+OPC Agent's core design philosophy centers on **Self-Evolution + Omnichannel + Role Templates + Production Toolchain**. As the **Agent Runtime** in The Self-Evolving Agent Stack, OPC focuses on how a single agent can **continuously self-evolve, reach users everywhere, and ship to production out of the box**.
+
+**🔄 Self-Evolution Flywheel**: Templates ship with Brain Seeds → Model Layer auto-learns on every call → Runtime runs continuously → Memory auto-evolves → Agent gets stronger over time.
+
+#### vs Hermes Agent
+
+**Design difference**: Hermes is a Python monolithic framework built around the Nous Research model ecosystem, emphasizing self-learning evolution and RL training. OPC is a TypeScript modular four-piece suite (Brain / Agent / Channel / CLI) emphasizing engineering productivity and out-of-box experience.
+
+**Where OPC leads**: Memory evolution trilogy (learn → recall → evolve with automatic clustering), 3-tier Brain Seed knowledge system, 25 built-in channels (vs Hermes's 16+), 100+ role templates, native A2A + AG-UI protocol support, OpenAI-compatible API server.
+
+**Where Hermes leads**: More mature Python ecosystem, battle-tested GRPO + LoRA RL training pipeline, larger Nous Research community, more mature Docker sandbox isolation.
+
+**Best for**: Choose Hermes if you need Python ecosystem and RL training capabilities. Choose OPC if you need TypeScript full-stack, memory evolution, omnichannel reach, and production toolchain.
+
+#### vs OpenClaw
+
+**Design difference**: OpenClaw is a heavyweight runtime with config-driven approach — install once, configure, and run. OPC is lightweight CLI + code-first, with four independent packages you can mix and match.
+
+**Where OPC leads**: Modular four-piece suite with independent packages, 100+ role templates, memory evolution system, 25 built-in channels, `opc init/dev/test/deploy` full-lifecycle CLI, A2A/AG-UI protocol support.
+
+**Where OpenClaw leads**: Polished out-of-box runtime experience, Puppeteer browser automation, mature permission and security sandbox, lower barrier to entry with config-driven approach.
+
+**Best for**: Choose OpenClaw if you prefer a "configure and run" all-in-one solution. Choose OPC if you prefer code-first, modular composition, and need memory evolution with role templates.
+
+#### vs CrewAI
+
+**Design difference**: CrewAI uses a Crew orchestration pattern — define agent roles, assign tasks, execute in sequence or parallel. OPC uses a Template + Memory model — each agent has independent memory and role templates, accumulating and evolving knowledge through DeepBrain.
+
+**Where OPC leads**: Memory evolution (not just short/long-term memory, but automatic clustering and distillation), Brain Seed pre-loaded knowledge, 25 built-in channels (CrewAI requires DIY), 100+ role templates, native TypeScript.
+
+**Where CrewAI leads**: Intuitive Crew + Flow orchestration model, Python ecosystem, 100K+ user mature community, feature-rich Enterprise edition.
+
+**Best for**: Choose CrewAI if you need multi-agent task orchestration and prefer Python. Choose OPC if you need deep single-agent evolution, omnichannel deployment, and TypeScript ecosystem.
+
+#### vs AutoGen
+
+**Design difference**: AutoGen is Microsoft's conversation-driven multi-agent framework, centered on agent-to-agent dialogue and UserProxy human-in-the-loop. OPC is task-driven with a template-based model, emphasizing single-agent memory evolution and production deployment.
+
+**Where OPC leads**: Memory evolution system, Brain Seed, 25 built-in channels, 100+ role templates, full-lifecycle CLI, YAML declarative configuration.
+
+**Where AutoGen leads**: Microsoft ecosystem support, distributed agent capabilities, AutoGen Studio visualization, mature UserProxy human-in-the-loop pattern, Python/C# dual-language support, active community.
+
+**Best for**: Choose AutoGen if you need multi-agent conversational collaboration and Microsoft ecosystem integration. Choose OPC if you need memory evolution, omnichannel reach, role templates, and TypeScript toolchain.
+
+#### Quick Reference Matrix
+
+| Feature | OPC Agent | Hermes Agent | OpenClaw | CrewAI | AutoGen |
+|---|:-:|:-:|:-:|:-:|:-:|
+| **Language** | TypeScript | Python | TypeScript | Python | Python/C# |
+| **Self-Evolution** | ✅ learn→recall→evolve | 🔶 | ❌ | ❌ | ❌ |
+| **Brain Seed** | ✅ 3-tier | ❌ | ❌ | ❌ | ❌ |
+| **Built-in Channels** | 25 | 16+ | 2 | ❌ | ❌ |
+| **Role Templates** | 100+ | ❌ | ❌ | ❌ | ❌ |
+| **Full-lifecycle CLI** | ✅ 20+ cmds | 🔶 | 🔶 | 🔶 | 🔶 |
+| **A2A + AG-UI** | ✅ | 🔶 / ❌ | ❌ | ❌ | ❌ |
+| **Multi-Agent** | ✅ | ✅ | 🔶 | ✅ | ✅ |
+| **RL Training** | ✅ Feedback opt. | ✅ GRPO+LoRA | ❌ | ❌ | ❌ |
+| **Browser Automation** | ✅ | ✅ | ✅ | ❌ | ❌ |
+| **Visual Dashboard** | ✅ OPC Studio | ✅ Dashboard | ❌ | 🔶 | ✅ Studio |
+| **Community Maturity** | 🚧 Early | ✅ Active | 🚧 Niche | ✅ Large | ✅ Large |
+| **License** | Apache-2.0 | MIT | MIT | Apache-2.0 | MIT |
+
+> Comparison based on each project's public documentation as of April 2026. All frameworks are evolving rapidly — corrections welcome via [Issues](https://github.com/Deepleaper/opc-agent/issues).
 
 ---
 
@@ -517,10 +827,10 @@ OPC Agent's differentiator is **Memory Evolution + Full-lifecycle CLI + multi-ch
 
 | Package | Purpose | Install |
 |---|---|---|
-| **[opc-agent](https://www.npmjs.com/package/opc-agent)** | Agent OS — create, run, manage | `npm i opc-agent` |
-| **[deepbrain](https://www.npmjs.com/package/deepbrain)** | Org Brain — memory storage & evolution | `npm i deepbrain` |
-| **[agentkits](https://www.npmjs.com/package/agentkits)** | Model Layer — unified API + routing | `npm i agentkits` |
-| **[agent-workstation](https://www.npmjs.com/package/agent-workstation)** | Role Templates — 100+ professional roles | `npm i agent-workstation` |
+| **[opc-agent](https://www.npmjs.com/package/opc-agent)** | Agent Runtime — create, run, manage | `npm i opc-agent` |
+| **[deepbrain](https://www.npmjs.com/package/deepbrain)** | Agent Memory — self-evolving memory engine | `npm i deepbrain` |
+| **[agentkits](https://www.npmjs.com/package/agentkits)** | Agent Model Layer — one-line LLM access with built-in memory | `npm i agentkits` |
+| **[agent-workstation](https://www.npmjs.com/package/agent-workstation)** | Agent Templates — 100+ professional role templates | `npm i agent-workstation` |
 
 ---
 
@@ -572,7 +882,7 @@ OPC Agent's differentiator is **Memory Evolution + Full-lifecycle CLI + multi-ch
 
 ---
 
-## 📡 11 Channels
+## 📡 25 Channels
 
 | Channel | Status | Description |
 |---------|:------:|-------------|
@@ -583,10 +893,24 @@ OPC Agent's differentiator is **Memory Evolution + Full-lifecycle CLI + multi-ch
 | 📧 Email | ✅ | IMAP / SMTP |
 | 💚 WeChat | ✅ | Enterprise / Personal WeChat |
 | 🔵 Feishu | ✅ | Feishu (Lark) Bot |
-| 🎤 Voice | ✅ | Voice call |
+| 🎤 Voice | ✅ | Voice call (TTS + STT) |
 | 🔌 WebSocket | ✅ | Real-time bidirectional |
 | 🪝 Webhook | ✅ | HTTP callback |
 | 📡 REST API | ✅ | HTTP API |
+| 💬 WhatsApp | ✅ | WhatsApp Business API |
+| 🟢 LINE | ✅ | LINE Messaging API |
+| 🟣 Teams | ✅ | Microsoft Teams Bot |
+| 📲 SMS | ✅ | Twilio / Cloud messaging |
+| 🔷 DingTalk | ✅ | DingTalk Bot |
+| 🐦 Twitter/X | ✅ | DM + Mentions |
+| 📸 Instagram | ✅ | Instagram DM |
+| 📘 Facebook | ✅ | Messenger API |
+| 🔐 Signal | ✅ | Signal Bot |
+| 🟩 Matrix | ✅ | Matrix protocol |
+| 💬 XMPP | ✅ | Jabber/XMPP |
+| 💻 IRC | ✅ | IRC protocol |
+| 📞 Twilio | ✅ | Phone / IVR |
+| 🔧 Custom | ✅ | Custom channel adapter |
 
 ---