OpenClaw Personal AI Assistant Complete Guide

OpenClaw 個人 AI 助手完全指南

Imagine having an AI assistant that runs 24/7 on your own machine, monitors your inbox, manages your calendar, syncs with your note-taking app, and responds through your favorite messaging platform — all while you sleep. That's OpenClaw.

This guide covers everything you need to know about OpenClaw: what it is, how it differs from Claude Code, what you need to set it up, and — critically — how to do so securely.

想像擁有一個 AI 助手，24/7 在你自己的機器上運行，監控你的收件匣、管理你的行事曆、同步你的筆記應用，並透過你喜歡的通訊平台回應——而且在你睡覺時也能運作。這就是 OpenClaw。

本指南涵蓋你需要知道的關於 OpenClaw 的一切：它是什麼、它與 Claude Code 有何不同、設置需要什麼，以及——至關重要的——如何安全地進行配置。

1. What is OpenClaw?

1. OpenClaw 是什麼？

OpenClaw is a free and open-source autonomous AI agent developed by Peter Steinberger (former PSPDFKit founder). It runs locally on your machine and connects to Large Language Models like Claude, GPT, or DeepSeek to execute tasks on your behalf.

A Brief History

Date	Name	Event
November 2025	Clawdbot	Original release by Peter Steinberger
January 27, 2026	Moltbot	Renamed after Anthropic trademark complaint
January 30, 2026	OpenClaw	Final rename ("Moltbot never quite rolled off the tongue")
February 2026	—	Over 100,000 GitHub stars, 5,700+ community skills on ClawHub

How It Works

Unlike cloud-based AI assistants, OpenClaw:

Runs locally — On your machine, VPS, or home server
Uses messaging as UI — Telegram, WhatsApp, Discord, Slack, Signal, or iMessage
Maintains persistent memory — Sessions span hours or days with context preservation
Connects to services via MCP — Model Context Protocol enables 100+ third-party integrations
Supports multiple LLMs — Claude, GPT-4, DeepSeek, and others

Configuration and interaction history are stored locally, enabling adaptive behavior across sessions.

OpenClaw 是由 Peter Steinberger（前 PSPDFKit 創辦人）開發的免費開源自主 AI 代理。它在你的本機上運行，並連接到 Claude、GPT 或 DeepSeek 等大型語言模型來代你執行任務。

簡史

日期	名稱	事件
2025 年 11 月	Clawdbot	Peter Steinberger 原始發布
2026 年 1 月 27 日	Moltbot	因 Anthropic 商標投訴而更名
2026 年 1 月 30 日	OpenClaw	最終更名（「Moltbot 念起來就是不順口」）
2026 年 2 月	—	超過 100,000 GitHub stars，ClawHub 上有 5,700+ 社群技能

運作方式

與雲端 AI 助手不同，OpenClaw：

本地運行 — 在你的機器、VPS 或家用伺服器上
以訊息為介面 — Telegram、WhatsApp、Discord、Slack、Signal 或 iMessage
維持持久記憶 — 會話可跨越數小時或數天並保留上下文
透過 MCP 連接服務 — Model Context Protocol 實現 100+ 第三方整合
支援多種 LLM — Claude、GPT-4、DeepSeek 等

配置和互動歷史儲存在本地，實現跨會話的適應性行為。

2. OpenClaw vs. Claude Code: Complementary, Not Competing

2. OpenClaw vs. Claude Code：互補而非競爭

A common misconception is that you must choose between OpenClaw and Claude Code. In reality, they serve fundamentally different purposes.

Aspect	Claude Code	OpenClaw
Primary domain	Software development	Personal automation
Interface	Terminal	Messaging apps
Session model	Per-task, temporary	Persistent daemon
Memory	Within conversation	Long-term, cross-session
Best for	Code refactoring, debugging, git workflows	Email, calendar, scheduled tasks, multi-platform

When to Use Claude Code

You're actively writing or refactoring code
You need deep codebase understanding
You want git-aware development workflows
You're debugging or writing tests

When to Use OpenClaw

You need 24/7 autonomous operation
Tasks span multiple sessions (monitoring inbox, follow-ups)
You want to interact via messaging platforms
You're automating non-coding workflows (calendar, smart home, health tracking)

Using Both Together

Many developers run both simultaneously:

Claude Code for focused coding sessions where code understanding matters
OpenClaw for everything else — reminders via WhatsApp, monitoring tasks, controlling home office setup

Since OpenClaw can use Claude as its AI backend, you get the same intelligence with different interfaces optimized for different use cases.

一個常見的誤解是你必須在 OpenClaw 和 Claude Code 之間做選擇。實際上，它們服務於根本不同的目的。

面向	Claude Code	OpenClaw
主要領域	軟體開發	個人自動化
介面	終端機	通訊 App
會話模式	按任務、暫時性	持久守護程序
記憶	對話內	長期、跨會話
最適合	重構程式碼、除錯、git 工作流	電郵、行事曆、排程任務、多平台

何時使用 Claude Code

你正在積極撰寫或重構程式碼
你需要深度理解程式碼庫
你想要 git 感知的開發工作流
你正在除錯或撰寫測試

何時使用 OpenClaw

你需要 24/7 自主運作
任務跨越多個會話（監控收件匣、跟進）
你想透過通訊平台互動
你正在自動化非程式碼工作流（行事曆、智慧家居、健康追蹤）

同時使用兩者

許多開發者同時運行兩者：

Claude Code 用於需要程式碼理解的專注程式碼會話
OpenClaw 用於其他一切——透過 WhatsApp 的提醒、監控任務、控制家庭辦公室設置

由於 OpenClaw 可以使用 Claude 作為其 AI 後端，你可以獲得相同的智慧，但有針對不同使用情境優化的不同介面。

3. System Requirements

3. 系統需求

Local Installation

Component	Requirement
Runtime	Node.js 22.12.0 or later
Package Manager	npm, pnpm, or bun
Platforms	macOS, Linux, Windows (WSL2 required)

Important: Native Windows is untested and has known compatibility issues. WSL2 is strongly recommended for Windows users.

VPS/Server Deployment (Recommended)

Component	Requirement
OS	Ubuntu 22.04 or 24.04
Access	Root or sudo privileges
Container	Docker and Docker Compose
RAM	Minimum 2GB (4GB recommended)
Disk	10GB+ for Docker images and logs

Additional Dependencies

build-essential package for native module compilation (Ubuntu/Debian)
API key for your chosen LLM provider (Anthropic, OpenAI, etc.)
Messaging platform account (Telegram, Discord, WhatsApp, etc.)

本機安裝

元件	需求
執行環境	Node.js 22.12.0 或更新版本
套件管理器	npm、pnpm 或 bun
平台	macOS、Linux、Windows（需要 WSL2）

**重要：**原生 Windows 未經測試且有已知的相容性問題。強烈建議 Windows 用戶使用 WSL2。

VPS/伺服器部署（推薦）

元件	需求
作業系統	Ubuntu 22.04 或 24.04
存取權限	Root 或 sudo 權限
容器	Docker 和 Docker Compose
RAM	最少 2GB（建議 4GB）
硬碟	10GB+ 用於 Docker 映像和日誌

額外依賴

build-essential 套件用於原生模組編譯（Ubuntu/Debian）
你選擇的 LLM 提供者的 API 金鑰（Anthropic、OpenAI 等）
通訊平台帳號（Telegram、Discord、WhatsApp 等）

4. Installation Methods

4. 安裝方法

Method 1: Quick Install (Easiest)

bash

# Install globally
npm install -g openclaw@latest

# Run onboarding wizard
openclaw onboard

# Start chatting via your chosen messaging platform

The onboarding wizard guides you through:

Selecting your AI provider (Claude, GPT-4, etc.)
Logging into your provider account
Setting up your messaging channel (Telegram, etc.)

Method 2: Docker Installation (Recommended for Security)

bash

# Clone the repository
git clone https://github.com/openclaw/openclaw.git
cd openclaw

# Run setup script
./setup.sh

Docker mounts two volumes:

~/.openclaw — Configuration and credentials
~/openclaw/workspace — Agent's sandbox environment

Method 3: Source Installation (For Contributors)

bash

# Clone repository
git clone https://github.com/openclaw/openclaw.git
cd openclaw

# Install dependencies
pnpm install

# Run development mode
pnpm dev

This path requires Git, Node.js, and pnpm, giving access to the latest development features.

方法一：快速安裝（最簡單）

bash

# 全域安裝
npm install -g openclaw@latest

# 執行引導精靈
openclaw onboard

# 透過你選擇的通訊平台開始對話

引導精靈會引導你完成：

選擇你的 AI 提供者（Claude、GPT-4 等）
登入你的提供者帳號
設置你的通訊頻道（Telegram 等）

方法二：Docker 安裝（安全性推薦）

bash

# 複製儲存庫
git clone https://github.com/openclaw/openclaw.git
cd openclaw

# 執行設置腳本
./setup.sh

Docker 掛載兩個卷：

~/.openclaw — 配置和憑證
~/openclaw/workspace — 代理的沙盒環境

方法三：原始碼安裝（適合貢獻者）

bash

# 複製儲存庫
git clone https://github.com/openclaw/openclaw.git
cd openclaw

# 安裝依賴
pnpm install

# 執行開發模式
pnpm dev

這個路徑需要 Git、Node.js 和 pnpm，可以存取最新的開發功能。

5. Setting Up Telegram as Your Interface

5. 設置 Telegram 作為你的介面

Telegram is one of the most popular interfaces for OpenClaw. Here's how to set it up:

Step 1: Create a Bot via BotFather

Open Telegram and search for @BotFather
Send /newbot to create a new bot
Follow prompts to set a name and username
Save the token — BotFather will send you a token like 123456789:ABCdefGHIjklMNOpqrsTUVwxyz

Security: If your token ever leaks, immediately use /revoke with BotFather to regenerate it.

Step 2: Configure OpenClaw

Add to your configuration file:

json

{
  "channels": {
    "telegram": {
      "enabled": true,
      "botToken": "YOUR_BOT_TOKEN_HERE",
      "dmPolicy": "pairing"
    }
  }
}

Step 3: Pair Your Account

By default, the bot isn't public — it requires pairing:

Send /start to your bot in Telegram
The bot returns a pairing code
Approve the pairing:

bash

# For Docker installations
docker compose run --rm openclaw-cli pairing approve telegram <CODE>

# For npm installations
openclaw pairing approve telegram <CODE>

Step 4: Get Your User ID (For Security)

Search for @userinfobot in Telegram
Click "Start" to see your numeric user ID
Add this ID to OpenClaw's allowed users list

Step 5: Configure Group Privacy (Optional)

If using OpenClaw in group chats:

Message @BotFather
Use /setprivacy
Select your bot
Choose "Disable" for full message access
Remove and re-add the bot to any existing groups

Note: With privacy mode ON, the bot only sees direct mentions and commands.

Telegram 是 OpenClaw 最受歡迎的介面之一。以下是設置方法：

步驟一：透過 BotFather 建立機器人

開啟 Telegram 並搜尋 @BotFather
發送 /newbot 建立新機器人
按提示設置名稱和用戶名
儲存 token — BotFather 會發送類似 123456789:ABCdefGHIjklMNOpqrsTUVwxyz 的 token

**安全：**如果你的 token 洩漏，立即使用 BotFather 的 /revoke 來重新產生。

步驟二：配置 OpenClaw

加入你的配置檔：

json

{
  "channels": {
    "telegram": {
      "enabled": true,
      "botToken": "你的_BOT_TOKEN",
      "dmPolicy": "pairing"
    }
  }
}

步驟三：配對你的帳號

預設情況下，機器人不是公開的——需要配對：

在 Telegram 中向你的機器人發送 /start
機器人返回配對碼
批准配對：

bash

# Docker 安裝
docker compose run --rm openclaw-cli pairing approve telegram <CODE>

# npm 安裝
openclaw pairing approve telegram <CODE>

步驟四：取得你的用戶 ID（安全用途）

在 Telegram 中搜尋 @userinfobot
點擊「Start」查看你的數字用戶 ID
將此 ID 加入 OpenClaw 的允許用戶清單

步驟五：配置群組隱私（可選）

如果在群組聊天中使用 OpenClaw：

私訊 @BotFather
使用 /setprivacy
選擇你的機器人
選擇「Disable」以獲得完整訊息存取
從現有群組中移除並重新添加機器人

**注意：**隱私模式開啟時，機器人只能看到直接提及和命令。

6. Integrating Services via MCP and Skills

6. 透過 MCP 和技能整合服務

OpenClaw's power comes from its extensible architecture. With 50+ built-in skills, 5,700+ community contributions on ClawHub, and MCP support for 100+ services, the possibilities are vast.

What Are Skills?

Skills are modular capabilities that extend OpenClaw. They're published as SKILL.md files with supporting code.

ClawHub: The Skill Registry

ClawHub is OpenClaw's public skill registry. You can:

Search for skills by plain language
Install skills with one command
Update skills easily
Publish your own skills

bash

# Install a skill
npx clawhub install <skill_name>

# Search for skills
npx clawhub search "gmail"

Key Integrations

Service	Capability
Gmail/Outlook	Read messages, draft replies, organize folders, auto-categorize, priority flagging
Google Calendar	Create events, check conflicts, send reminders
Todoist/Things/Asana	Add tasks, check lists, mark complete
Obsidian/Notion	Add notes, search vault, bidirectional sync
Slack	Summarize channels, draft messages, search history, respond to mentions
GitHub	Check PRs, review issues, monitor notifications
WHOOP	Daily fitness summaries
Philips Hue	Smart home control
Spotify	Playback control

Obsidian Integration Details

The Obsidian skill works with your local vault (plain Markdown files):

Vault Detection: OpenClaw reads from ~/Library/Application Support/obsidian/obsidian.json (macOS)
Best Practice: Don't hardcode vault paths — let the skill detect them
Capabilities: Add notes, search vault, retrieve existing notes

bash

# Install Obsidian skill
npx clawhub install obsidian

Configure your vault path in TOOLS.md so OpenClaw knows where to save responses.

MCP Server Configuration

For advanced integrations, configure MCP servers in JSON:

json

{
  "mcpServers": {
    "gmail": {
      "command": "npx",
      "args": ["@anthropic/mcp-server-gmail"],
      "env": {
        "GMAIL_CREDENTIALS": "/path/to/credentials.json"
      }
    }
  }
}

OpenClaw 的強大來自其可擴展的架構。有 50+ 內建技能、ClawHub 上 5,700+ 社群貢獻，以及支援 100+ 服務的 MCP，可能性是巨大的。

什麼是技能？

技能是擴展 OpenClaw 的模組化能力。它們以 SKILL.md 檔案和支援程式碼的形式發布。

ClawHub：技能登錄處

ClawHub 是 OpenClaw 的公開技能登錄處。你可以：

搜尋技能（用自然語言）
安裝技能（一個命令）
更新技能（輕鬆完成）
發布你自己的技能

bash

# 安裝技能
npx clawhub install <skill_name>

# 搜尋技能
npx clawhub search "gmail"

主要整合

服務	能力
Gmail/Outlook	閱讀訊息、草擬回覆、整理資料夾、自動分類、優先標記
Google Calendar	建立事件、檢查衝突、發送提醒
Todoist/Things/Asana	新增任務、檢查清單、標記完成
Obsidian/Notion	新增筆記、搜尋 vault、雙向同步
Slack	摘要頻道、草擬訊息、搜尋歷史、回應提及
GitHub	檢查 PR、審查 issue、監控通知
WHOOP	每日健身摘要
Philips Hue	智慧家居控制
Spotify	播放控制

Obsidian 整合細節

Obsidian 技能與你的本地 vault（純 Markdown 檔案）配合使用：

**Vault 偵測：**OpenClaw 從 ~/Library/Application Support/obsidian/obsidian.json（macOS）讀取
**最佳實踐：**不要硬編碼 vault 路徑——讓技能自動偵測
**能力：**新增筆記、搜尋 vault、檢索現有筆記

bash

# 安裝 Obsidian 技能
npx clawhub install obsidian

在 TOOLS.md 中配置你的 vault 路徑，讓 OpenClaw 知道該在哪裡儲存回應。

MCP 伺服器配置

對於進階整合，在 JSON 中配置 MCP 伺服器：

json

{
  "mcpServers": {
    "gmail": {
      "command": "npx",
      "args": ["@anthropic/mcp-server-gmail"],
      "env": {
        "GMAIL_CREDENTIALS": "/path/to/credentials.json"
      }
    }
  }
}

7. Scheduling with Cron Jobs

7. 使用 Cron 排程

One of OpenClaw's killer features is scheduled autonomous tasks:

json

{
  "cron": {
    "dailyBriefing": {
      "schedule": "0 7 * * *",
      "task": "Send me a summary of today's calendar and top 3 priority emails"
    },
    "eveningTodoReview": {
      "schedule": "0 22 * * *",
      "task": "Draft tomorrow's to-do list based on calendar and pending tasks"
    },
    "hourlyInboxCheck": {
      "schedule": "0 * * * *",
      "task": "Check inbox and flag anything urgent"
    },
    "weeklyReport": {
      "schedule": "0 9 * * 0",
      "task": "Compile weekly report from completed tasks and sent emails"
    }
  }
}

You set it once. It runs forever. No manual triggering required.

OpenClaw 的殺手級功能之一是排程自主任務：

json

{
  "cron": {
    "dailyBriefing": {
      "schedule": "0 7 * * *",
      "task": "發送今天的行事曆摘要和前 3 封優先郵件"
    },
    "eveningTodoReview": {
      "schedule": "0 22 * * *",
      "task": "根據行事曆和待辦任務草擬明天的待辦清單"
    },
    "hourlyInboxCheck": {
      "schedule": "0 * * * *",
      "task": "檢查收件匣並標記任何緊急事項"
    },
    "weeklyReport": {
      "schedule": "0 9 * * 0",
      "task": "從已完成任務和已發送郵件編譯週報"
    }
  }
}

你設置一次。它永遠運行。不需要手動觸發。

8. Security: The Critical Consideration

8. 安全性：關鍵考量

This is the most important section of this guide.

OpenClaw is powerful precisely because it can execute real actions on your behalf — shell commands, file operations, network requests. This makes security non-negotiable.

Known Risks

Sandbox Mode is OFF by Default
- Direct message sessions aren't isolated by default
- Public agents can be created with broad access unless carefully constrained
Malicious Skills
- ClawHub hosts 5,700+ skills — not all are trustworthy
- VirusTotal has flagged hundreds of skills with malicious properties
- Always review skill source code before installing
Prompt Injection
- AI models cannot reliably distinguish legitimate commands from malicious instructions embedded in data
- This is not theoretical — it's documented and reproducible
Exposed Deployments
- Many users expose admin dashboards with weak authentication
- Email, cloud storage, and calendar access becomes attack surface

Security Best Practices

1. Enable Sandbox Mode

json

{
  "sandbox": {
    "mode": "all",
    "workspaceAccess": "none",
    "docker": {
      "network": "none"
    }
  }
}

2. Use Docker Isolation

Never run OpenClaw directly on your host system. Docker provides:

Isolated filesystem
Restricted network access
CPU and memory limits
Limited blast radius if compromised

3. Create a Dedicated User

bash

# Create dedicated user
sudo useradd -m openclaw

# Run OpenClaw as this user
sudo -u openclaw openclaw start

4. Limit High-Risk Tools

Deny by default, allow explicitly:

json

{
  "tools": {
    "deny": ["exec", "browser", "web_fetch", "gateway", "nodes", "cron"],
    "allow": ["calendar", "email_read", "notes"]
  }
}

5. Review Skills Before Installing

Check VirusTotal reports on ClawHub skill pages
Read the source code
Use Claude Code or similar to audit for malicious behavior

6. Deploy on Isolated Infrastructure

Instead of your personal machine:

Use a dedicated VPS or home server
Run in a VM with network isolation
Use firewall rules to limit outbound connections

The Security Mindset

"OpenClaw must ONLY be operated in a fully isolated sandbox environment. Even if the agent gets compromised, it cannot cause damage outside the sandbox."

If you wouldn't give a stranger the credentials you're giving OpenClaw, you shouldn't give them to OpenClaw either — unless it's running in a properly sandboxed environment.

這是本指南最重要的部分。

OpenClaw 之所以強大，正是因為它可以代你執行真實操作——shell 命令、檔案操作、網路請求。這使得安全性不容妥協。

已知風險

沙盒模式預設關閉
- 直接訊息會話預設沒有隔離
- 除非仔細約束，否則可以建立具有廣泛存取權限的公開代理
惡意技能
- ClawHub 託管 5,700+ 技能——並非全部可信
- VirusTotal 已標記數百個具有惡意屬性的技能
- 安裝前務必審查技能原始碼
提示注入
- AI 模型無法可靠地區分嵌入在資料中的合法命令和惡意指令
- 這不是理論性的——它是有記錄且可重現的
暴露的部署
- 許多用戶使用弱驗證暴露管理儀表板
- 電子郵件、雲端儲存和行事曆存取成為攻擊面

安全最佳實踐

1. 啟用沙盒模式

json

{
  "sandbox": {
    "mode": "all",
    "workspaceAccess": "none",
    "docker": {
      "network": "none"
    }
  }
}

2. 使用 Docker 隔離

永遠不要直接在主機系統上運行 OpenClaw。Docker 提供：

隔離的檔案系統
受限的網路存取
CPU 和記憶體限制
如果被入侵，爆炸半徑有限

3. 建立專用用戶

bash

# 建立專用用戶
sudo useradd -m openclaw

# 以此用戶運行 OpenClaw
sudo -u openclaw openclaw start

4. 限制高風險工具

預設拒絕，明確允許：

json

{
  "tools": {
    "deny": ["exec", "browser", "web_fetch", "gateway", "nodes", "cron"],
    "allow": ["calendar", "email_read", "notes"]
  }
}

5. 安裝前審查技能

在 ClawHub 技能頁面檢查 VirusTotal 報告
閱讀原始碼
使用 Claude Code 或類似工具審計惡意行為

6. 部署在隔離基礎設施上

不要用你的個人機器：

使用專用 VPS 或家用伺服器
在具有網路隔離的 VM 中運行
使用防火牆規則限制出站連接

安全心態

「OpenClaw 必須只在完全隔離的沙盒環境中運作。即使代理被入侵，它也無法對沙盒外造成損害。」

如果你不會把你給 OpenClaw 的憑證給陌生人，你也不應該給 OpenClaw——除非它運行在正確沙盒化的環境中。

9. Troubleshooting Common Issues

9. 常見問題疑難排解

Installation Issues

Problem: npm install fails or "command not found"

Solution:

Ensure Node.js 22.12.0+ is installed
Restart terminal after installation
Add npm global bin to PATH:

bash

export PATH="$PATH:$(npm config get prefix)/bin"

Telegram Connection Issues

Problem: Bot starts then silently stops responding

Solution:

Some hosts resolve api.telegram.org to IPv6 first
If your server lacks IPv6 egress, force IPv4 or enable IPv6

Problem: HttpError: Network request failed

Solution:

Check firewall rules for outbound HTTPS
Verify API key is correct and has credits

Onboarding Issues

Problem: Wizard exits or channel linking fails

Solution:

Run openclaw onboard again
You can redo just the channel step

Gateway Issues

Problem: Persistent gateway errors

Solution:

bash

openclaw doctor --fix

Missing API Key

Problem: OpenClaw won't work at all

Solution:

The most common config mistake is a missing API key
Verify you have a valid Anthropic or OpenAI API key configured

安裝問題

問題：npm install 失敗或「command not found」

解決方案：

確保安裝了 Node.js 22.12.0+
安裝後重啟終端機
將 npm 全域 bin 加入 PATH：

bash

export PATH="$PATH:$(npm config get prefix)/bin"

Telegram 連接問題

**問題：**機器人啟動後靜默停止回應

解決方案：

某些主機首先將 api.telegram.org 解析為 IPv6
如果你的伺服器缺少 IPv6 出口，強制使用 IPv4 或啟用 IPv6

**問題：**HttpError: Network request failed

解決方案：

檢查出站 HTTPS 的防火牆規則
驗證 API 金鑰正確且有餘額

引導問題

**問題：**精靈退出或頻道連結失敗

解決方案：

再次運行 openclaw onboard
你可以只重做頻道步驟

閘道問題

**問題：**持續的閘道錯誤

解決方案：

bash

openclaw doctor --fix

缺少 API 金鑰

**問題：**OpenClaw 完全無法運作

解決方案：

最常見的配置錯誤是缺少 API 金鑰
驗證你已配置有效的 Anthropic 或 OpenAI API 金鑰

10. Complete Setup Checklist

10. 完整設置清單

Here's everything you need to achieve the setup you described (Gmail, Todoist, Slack, Obsidian, local operation, Telegram interface):

Prerequisites

[ ] VPS or dedicated machine (not your personal computer)
[ ] Node.js 22.12.0+ installed
[ ] Docker and Docker Compose installed
[ ] Anthropic API key (or OpenAI/DeepSeek)
[ ] Telegram account

Installation

[ ] Install OpenClaw via Docker (recommended) or npm
[ ] Run openclaw onboard wizard
[ ] Select Claude as AI provider
[ ] Configure Telegram as messaging channel

Telegram Setup

[ ] Create bot via @BotFather
[ ] Save bot token securely
[ ] Configure token in OpenClaw
[ ] Pair your Telegram account
[ ] Get your user ID via @userinfobot
[ ] Add user ID to allowed list

Security Configuration

[ ] Enable sandbox mode
[ ] Create dedicated Linux user for OpenClaw
[ ] Configure Docker network isolation
[ ] Set tool allowlists (deny high-risk by default)
[ ] Review any skills before installing

Service Integrations

[ ] Install Gmail skill: npx clawhub install gmail
[ ] Configure Gmail OAuth credentials
[ ] Install Todoist skill: npx clawhub install todoist
[ ] Get Todoist API token and configure
[ ] Install Slack skill: npx clawhub install slack
[ ] Create Slack app and configure OAuth
[ ] Install Obsidian skill: npx clawhub install obsidian
[ ] Configure vault path in TOOLS.md

Automation

[ ] Set up cron jobs for scheduled tasks
[ ] Test each integration individually
[ ] Monitor logs for the first few days

Ongoing

[ ] Regularly update OpenClaw: npm update -g openclaw
[ ] Review ClawHub security reports for installed skills
[ ] Monitor API usage and costs

以下是達成你描述的設置（Gmail、Todoist、Slack、Obsidian、本機運作、Telegram 介面）所需的一切：

先決條件

[ ] VPS 或專用機器（不是你的個人電腦）
[ ] 已安裝 Node.js 22.12.0+
[ ] 已安裝 Docker 和 Docker Compose
[ ] Anthropic API 金鑰（或 OpenAI/DeepSeek）
[ ] Telegram 帳號

安裝

[ ] 透過 Docker（推薦）或 npm 安裝 OpenClaw
[ ] 運行 openclaw onboard 精靈
[ ] 選擇 Claude 作為 AI 提供者
[ ] 配置 Telegram 作為通訊頻道

Telegram 設置

[ ] 透過 @BotFather 建立機器人
[ ] 安全儲存機器人 token
[ ] 在 OpenClaw 中配置 token
[ ] 配對你的 Telegram 帳號
[ ] 透過 @userinfobot 取得你的用戶 ID
[ ] 將用戶 ID 加入允許清單

安全配置

[ ] 啟用沙盒模式
[ ] 為 OpenClaw 建立專用 Linux 用戶
[ ] 配置 Docker 網路隔離
[ ] 設定工具允許清單（預設拒絕高風險）
[ ] 安裝前審查任何技能

服務整合

[ ] 安裝 Gmail 技能：npx clawhub install gmail
[ ] 配置 Gmail OAuth 憑證
[ ] 安裝 Todoist 技能：npx clawhub install todoist
[ ] 取得 Todoist API token 並配置
[ ] 安裝 Slack 技能：npx clawhub install slack
[ ] 建立 Slack 應用並配置 OAuth
[ ] 安裝 Obsidian 技能：npx clawhub install obsidian
[ ] 在 TOOLS.md 中配置 vault 路徑

自動化

[ ] 為排程任務設置 cron 作業
[ ] 單獨測試每個整合
[ ] 最初幾天監控日誌

持續維護

[ ] 定期更新 OpenClaw：npm update -g openclaw
[ ] 審查已安裝技能的 ClawHub 安全報告
[ ] 監控 API 使用量和成本

Conclusion: The Promise and the Price

OpenClaw represents a genuine leap in personal AI assistants — persistent, autonomous, locally-run, and deeply integrated with your digital life. The ability to have an AI that monitors your inbox, manages your calendar, syncs with your notes, and responds via messaging — all while you sleep — is genuinely transformative.

But this power comes with responsibility:

Security is non-negotiable. Never run OpenClaw unsandboxed on a machine with sensitive data.
Review before you trust. Skills can be malicious. Audit them.
Isolate the blast radius. Use Docker, dedicated users, and network restrictions.
Start small. Enable one integration at a time. Test thoroughly.

Done right, OpenClaw can genuinely feel like having a tireless chief of staff. Done wrong, it's a security nightmare waiting to happen.

The choice — and the responsibility — is yours.

Sources:

結語：承諾與代價

OpenClaw 代表了個人 AI 助手的真正飛躍——持久、自主、本地運行，並與你的數位生活深度整合。擁有一個 AI 可以監控你的收件匣、管理你的行事曆、同步你的筆記，並透過訊息回應——而且在你睡覺時也能運作——這確實是變革性的。

但這種力量伴隨著責任：

**安全性不容妥協。**永遠不要在有敏感資料的機器上未經沙盒化運行 OpenClaw。
**信任前先審查。**技能可能是惡意的。審計它們。
**隔離爆炸半徑。**使用 Docker、專用用戶和網路限制。
**從小處開始。**一次啟用一個整合。徹底測試。

做對了，OpenClaw 真的可以感覺像是擁有一個不知疲倦的幕僚長。做錯了，它是一個等待發生的安全噩夢。

選擇——和責任——在你。

來源：

OpenClaw Personal AI Assistant Complete Guide ​

OpenClaw 個人 AI 助手完全指南 ​

1. What is OpenClaw? ​

1. OpenClaw 是什麼？ ​

A Brief History ​

How It Works ​

簡史 ​

運作方式 ​

2. OpenClaw vs. Claude Code: Complementary, Not Competing ​

2. OpenClaw vs. Claude Code：互補而非競爭 ​

When to Use Claude Code ​

When to Use OpenClaw ​

Using Both Together ​

何時使用 Claude Code ​

何時使用 OpenClaw ​

同時使用兩者 ​

3. System Requirements ​

3. 系統需求 ​

Local Installation ​

VPS/Server Deployment (Recommended) ​

Additional Dependencies ​

本機安裝 ​

VPS/伺服器部署（推薦） ​

額外依賴 ​

4. Installation Methods ​

4. 安裝方法 ​

Method 1: Quick Install (Easiest) ​

Method 2: Docker Installation (Recommended for Security) ​

Method 3: Source Installation (For Contributors) ​

方法一：快速安裝（最簡單） ​

方法二：Docker 安裝（安全性推薦） ​

方法三：原始碼安裝（適合貢獻者） ​

5. Setting Up Telegram as Your Interface ​

5. 設置 Telegram 作為你的介面 ​

Step 1: Create a Bot via BotFather ​

Step 2: Configure OpenClaw ​

Step 3: Pair Your Account ​

Step 4: Get Your User ID (For Security) ​

Step 5: Configure Group Privacy (Optional) ​

步驟一：透過 BotFather 建立機器人 ​

步驟二：配置 OpenClaw ​

步驟三：配對你的帳號 ​

步驟四：取得你的用戶 ID（安全用途） ​

步驟五：配置群組隱私（可選） ​

6. Integrating Services via MCP and Skills ​

6. 透過 MCP 和技能整合服務 ​

What Are Skills? ​

ClawHub: The Skill Registry ​

Key Integrations ​

Obsidian Integration Details ​

MCP Server Configuration ​

什麼是技能？ ​

ClawHub：技能登錄處 ​

主要整合 ​

Obsidian 整合細節 ​

MCP 伺服器配置 ​

7. Scheduling with Cron Jobs ​

7. 使用 Cron 排程 ​

8. Security: The Critical Consideration ​

8. 安全性：關鍵考量 ​

Known Risks ​

Security Best Practices ​

1. Enable Sandbox Mode ​

2. Use Docker Isolation ​

3. Create a Dedicated User ​

4. Limit High-Risk Tools ​

5. Review Skills Before Installing ​

6. Deploy on Isolated Infrastructure ​

The Security Mindset ​

已知風險 ​

安全最佳實踐 ​

1. 啟用沙盒模式 ​

2. 使用 Docker 隔離 ​

3. 建立專用用戶 ​

4. 限制高風險工具 ​

5. 安裝前審查技能 ​

6. 部署在隔離基礎設施上 ​

安全心態 ​

9. Troubleshooting Common Issues ​

9. 常見問題疑難排解 ​

OpenClaw Personal AI Assistant Complete Guide

OpenClaw 個人 AI 助手完全指南

1. What is OpenClaw?

1. OpenClaw 是什麼？

A Brief History

How It Works

簡史

運作方式

2. OpenClaw vs. Claude Code: Complementary, Not Competing

2. OpenClaw vs. Claude Code：互補而非競爭

When to Use Claude Code

When to Use OpenClaw

Using Both Together

何時使用 Claude Code

何時使用 OpenClaw

同時使用兩者

3. System Requirements

3. 系統需求

Local Installation

VPS/Server Deployment (Recommended)

Additional Dependencies

本機安裝

VPS/伺服器部署（推薦）

額外依賴

4. Installation Methods

4. 安裝方法

Method 1: Quick Install (Easiest)

Method 2: Docker Installation (Recommended for Security)

Method 3: Source Installation (For Contributors)

方法一：快速安裝（最簡單）

方法二：Docker 安裝（安全性推薦）

方法三：原始碼安裝（適合貢獻者）

5. Setting Up Telegram as Your Interface

5. 設置 Telegram 作為你的介面

Step 1: Create a Bot via BotFather

Step 2: Configure OpenClaw

Step 3: Pair Your Account

Step 4: Get Your User ID (For Security)

Step 5: Configure Group Privacy (Optional)

步驟一：透過 BotFather 建立機器人

步驟二：配置 OpenClaw

步驟三：配對你的帳號

步驟四：取得你的用戶 ID（安全用途）

步驟五：配置群組隱私（可選）

6. Integrating Services via MCP and Skills

6. 透過 MCP 和技能整合服務

What Are Skills?

ClawHub: The Skill Registry

Key Integrations

Obsidian Integration Details

MCP Server Configuration

什麼是技能？

ClawHub：技能登錄處

主要整合

Obsidian 整合細節

MCP 伺服器配置

7. Scheduling with Cron Jobs

7. 使用 Cron 排程

8. Security: The Critical Consideration

8. 安全性：關鍵考量

Known Risks

Security Best Practices

1. Enable Sandbox Mode

2. Use Docker Isolation

3. Create a Dedicated User

4. Limit High-Risk Tools

5. Review Skills Before Installing

6. Deploy on Isolated Infrastructure

The Security Mindset

已知風險

安全最佳實踐

1. 啟用沙盒模式

2. 使用 Docker 隔離

3. 建立專用用戶

4. 限制高風險工具

5. 安裝前審查技能

6. 部署在隔離基礎設施上

安全心態

9. Troubleshooting Common Issues

9. 常見問題疑難排解