Articles / Clawdbot Tutorial and Security Warning

Clawdbot Tutorial and Security Warning

29 1 月, 2026 • 4 min read • AI-agentslocal-AI-security

Clawdbot Tutorial and Security Warning

In recent days, a new AI agent — Clawdbot — has gone viral across tech communities. Known affectionately (and humorously) as the “chubby lobster” 🦞, this local-first, highly autonomous AI assistant promises unprecedented control over your personal computing environment.

⚠️ Important Note: As of early 2026, Clawdbot has been officially rebranded to Moltbot, following a trademark dispute with Anthropic. The name change was mandated due to market confusion with Claude Code — Anthropic claimed the original name implied official affiliation or extension.

Clawdbot mascot: chubby lobster

Why It’s Exploding — And Why You Should Pause

Clawdbot’s surge in popularity has triggered unexpected real-world effects — including Mac Mini shortages, with listings vanishing even on platforms like Meituan外卖. Why? Because Clawdbot runs locally, requires high system permissions, and operates with extreme proactivity — enabling actions like:

  • Editing local files and folders
  • Automating stock trading
  • Processing emails
  • Managing calendar & notifications
  • Executing terminal commands (e.g., rm -rf / — yes, seriously)

Mac Mini shortage screenshot

Its official tagline? “A personal AI assistant you run on your own device.”

Key Differentiators vs. Claude Code

Feature Clawdbot (Moltbot) Claude Code
Deployment Fully local, cross-platform (macOS/Windows/Linux) Cloud-first, tightly coupled with Anthropic infrastructure
Messaging Integration Native support for WhatsApp, Telegram, Discord; community-built bridges for Feishu, QQ, and Lark Limited to Anthropic’s official web UI and mobile app
Memory Model Persistent, file-based long-term memory stored locally — enables rich cross-session context Session-bound; no persistent memory outside Anthropic’s ecosystem
Extensibility Open-source, modular Skills architecture, model-agnostic (supports Qwen, GLM, Claude, Codex, etc.) Closed, proprietary, model-locked

Moltbot rebrand notice

Installation Guide — With Critical Safety Notes

⚠️ Before You Begin: Never install on your primary machine. Use a dedicated test device, virtual machine, or cloud instance (e.g., Tencent Cloud’s one-click Moltbot service).

Prerequisites

  • Node.js v22+
  • Git (for --install-method git)

Install Commands

# Windows (PowerShell)
iwr -useb https://molt.bot/install.ps1 | iex

# macOS / Linux / Ubuntu / Debian
curl -fsSL https://molt.bot/install.sh | bash -s -- --install-method git

Installation in progress

After installation, you’ll see a stark confirmation prompt:

“I understand this lobster is powerful — and dangerous. Do you accept the risk?”

✅ Select Yes to proceed — No terminates the process immediately.

Risk acknowledgment screen

Then choose Quick Start, followed by model configuration. Popular options include:

  • ✅ Codex OAuth (via ChatGPT Pro — safe & endorsed by OpenAI)
  • ❌ Avoid Claude Max API keys — Anthropic actively bans accounts using them outside Claude Code
  • ✅ Domestic models: Qwen, GLM-4.7, MiniMax (but expect aggressive token consumption — e.g., ~30M tokens in 48h with GLM-4.7)

Model selection screen

Post-Install Setup

  • Skip external chat integrations (WhatsApp/Telegram/Discord) if unused — select “Skip”
  • Enable recommended hooks:
  • boot-md: Load default instructions/rules at startup
  • command-logger: Log executed commands (opt-in for debugging)
  • session-memory: Retain conversational state across sessions
  • Install Skills selectively — or skip and add later via CLI/UI

Skills selection interface

Launch with:

clawdbot gateway --verbose

→ Access UI at: http://127.0.0.1:18789/chat

Moltbot Web UI

Integrating with Feishu (Lark) — A Domestic Workflow

Since native messaging support is limited in China, the community-developed Clawdbot-feishu bridge offers seamless integration.

Smart Self-Configuration (Yes, It Installs Itself)

From the Web UI, simply type:

Install Clawdbot plugin @m1heng-clawd/feishu

✅ Moltbot will auto-install the Feishu connector — saving hours of manual debugging.

Then provide your Feishu App ID and App Secret directly in-chat. Moltbot parses and configures them automatically.

Feishu app setup

Next, complete Feishu Open Platform setup:
– Add bot to group/chat
– Grant permissions: chat:read, message:send, contact:user:readonly
– Configure event subscriptions (message, url_verification)
– Publish the bot

Feishu permissions panel

Once live, message the bot in Feishu to trigger local automation — e.g., “Fill invoice data into template.xlsx” → Moltbot opens Excel, reads PDFs, populates cells, saves.

Feishu interaction demo

Final Reflection: Curiosity vs. Control

Clawdbot isn’t just software — it’s a philosophical mirror. It embodies our collective desire for AI that acts, not just answers. But autonomy demands accountability.

Think of it like hiring a super-capable but unvetted personal assistant:
– You might grant access to your kitchen (files), living room (browser), and office (email)
– But would you hand them keys to your bedroom (password vault, banking apps, private photos)?

The real question isn’t “Can it do this?” — it’s “Which doors do I want it to open — and which ones must remain locked?”

As GitHub stars soar past 63,000, and deployments scale globally, one truth remains: The most powerful feature of any AI agent isn’t its code — it’s your judgment.

GitHub star growth chart

Article originally published by Digital Life Kazek, authors Kazek & Lin Jimengdou.