Critical Warning: Understand the Risks Before You Begin

Clawdbot (now renamed Moltbot after legal pressure from Anthropic) is a locally-running AI agent with extraordinary power—and proportional danger. Unlike passive chatbots, Clawdbot has high-level system permissions allowing it to modify local files, manage emails, execute trades, and perform virtually any computer operation autonomously through messaging apps like Feishu, WhatsApp, or Telegram. While this “ultimate proactive agent” has driven Mac mini sales to unprecedented levels and earned 63,000+ GitHub stars, the security risks are severe: one careless command could result in deleted files, embarrassing social media posts, or drained bank accounts. This comprehensive tutorial will walk you through installation on Windows, Mac, and Linux, plus integration with Feishu for Chinese users—but the most important advice is: never install on your primary computer. Use a spare device, virtual machine, or cloud server. Deploy responsibly, understand what permissions you're granting, and remember: giving AI the keys to your digital life requires the same careful judgment as hiring a household manager.

The Clawdbot Phenomenon and Name Change

From Clawdbot to Moltbot

Original Name: Clawdbot (featuring a chubby lobster mascot 🦞)

Legal Action: Anthropic forced a name change, claiming “Clawdbot” could be misunderstood as an official Claude Code extension product

New Name: Moltbot (same product, different branding)

Community Reaction: Frustration with Anthropic's aggressive legal stance against open-source innovation

The Mac Mini Craze

Clawdbot's explosion created an unexpected hardware shortage:

Sales Impact:

• Mac mini units selling out rapidly
• Users purchasing specifically for Clawdbot deployment
• Even food delivery platforms showing “out of stock” for Mac mini
• Apple likely experiencing Q1 sales boost

Why Mac Mini?:

• Ideal 24/7 operation device
• Low power consumption
• Quiet operation for home deployment
• Sufficient processing power
• Convenient plug-and-play experience

Cost Reality: Mac mini represents $600-800 investment—not necessary but convenient for non-technical users

Understanding Clawdbot's Power and Danger

Official Definition

“A personal AI assistant running on your own device.”

Capability Overview

What Makes It Different from Claude Code:

• Claude Code suffers from name confusion (perceived as programming-only)
• Clawdbot explicitly positions as general-purpose agent
• Both have similar underlying capabilities
• Modern internet society built on code enables broad automation

Core Capabilities:

• Modify local files directly
• Execute stock trades
• Process emails automatically
• Control browsers and applications
• Run system commands with high privileges
• Access messaging platforms for remote control

The Security Nightmare Scenario

Real Risk Examples:

Accidental Social Media Disaster: Casual remark interpreted as command → embarrassing post shared publicly

File Deletion: Misunderstood instruction → critical documents permanently deleted

Financial Loss: Ambiguous order → unauthorized purchases or transfers

Real-World Case: User reported Clawdbot deleting files unexpectedly due to prompt misinterpretation

Why This Matters: These aren't theoretical risks—they're documented problems from actual users

Two Safe Deployment Approaches

Option 1: Dedicated Hardware

• Purchase spare computer for Clawdbot exclusively
• Use old laptop no longer needed for primary work
• Isolate completely from important data

Option 2: Cloud/Virtual Machine

• Deploy on cloud server (higher technical barrier)
• Use virtual machine for isolation
• Tencent Cloud now offers one-click deployment service

Mac Mini as Simple Solution:

• Most straightforward for non-technical users
• Energy efficient for 24/7 operation
• Complete isolation from primary systems
• Expensive but eliminates configuration complexity

Critical Advice: If you don't care deeply about security or insist on using your primary computer, ANY old computer from years ago can run Clawdbot—no need to “pay Apple tax” unnecessarily

Clawdbot vs. Claude Code: Key Differences

Three Major Distinctions

1. Messaging Platform Integration

Connects to international platforms:

• WhatsApp
• Telegram
• Discord

Chinese platform integrations (community-developed):

• Feishu (official solution emerging)
• QQ (experimental implementations in progress)

User Experience: Send message from phone → Clawdbot executes on local computer → seamless mobile-to-desktop control

2. Long-Term Memory

Storage Method: Memories saved as files locally

Functionality:

• Similar to ChatGPT's memory feature
• Retrieves relevant context from past conversations
• Cross-session accumulation
• Controllable and transparent
• Easy to understand and modify

Advantage: Memory persistence enables true personalization over time

3. Open Source Flexibility

Customization Freedom:

• Self-deployment on any infrastructure
• Connect to preferred AI models
• Similar to OpenCode's approach
• Supports Skills ecosystem

Model Agnostic: Use Claude, GPT, MiniMax, Qwen, GLM, or any compatible model

The Perfect Storm

Factors Driving Explosive Growth:

• Right timing (post-Claude Code excitement)
• Geographic advantages (fills gaps in Chinese market)
• Community momentum
• Anxiety about AI coding capabilities driving exploration
• Open-source nature enabling rapid iteration

Result: 63,000+ GitHub stars with “insanely steep” growth curve

Complete Installation Guide

Prerequisites

Essential Requirement: Node.js version 22 or higher

Download: https://nodejs.org (LTS version recommended)

Verification: After installation, run node --version to confirm

Installation Commands

Windows:

iwr -useb https://molt.bot/install.ps1 | iex

macOS / Linux / Ubuntu / Debian:

curl -fsSL https://molt.bot/install.sh | bash -s -- --install-method git

Critical Pre-Installation Warning

BEFORE RUNNING INSTALLATION:

⚠️ EXTREME WARNING: This tutorial uses a test laptop at the office. DO NOT install on your primary computer unless absolutely necessary.

⚠️ USE SPARE DEVICE: Always use backup machine for Clawdbot deployment

⚠️ HIGH PERMISSIONS: Clawdbot requires extensive system access

⚠️ POTENTIAL DISASTER: If AI malfunctions and executes rm -rf / command, all your private files and “study materials” will be permanently deleted

⚠️ NO REGRETS: Don't blame me if important files get deleted after ignoring this warning

Installation Process Walkthrough

Step 1: Acknowledge Risks

After installation command completes, you'll see this prompt:

“I understand this powerful lobster has great capabilities but extreme risks. Do you confirm and continue?”

Options:

• YES: Proceed with installation
• NO: Process terminates immediately (quite dictatorial!)

Only selecting YES allows you to continue—this mandatory acknowledgment forces risk awareness.

Step 2: Choose Setup Mode

Two options appear:

Option 1 – QuickStart:

• Fast initial setup
• Configure details later via clawdbot configure
• Recommended for most users

Option 2 – Manual:

• Configure everything during installation
• More control over initial settings
• Better for experienced users

Recommendation: Choose QuickStart for simplicity

Step 3: Model Configuration

System prompts for AI model selection.

Available Options:

• Claude (via Anthropic API)
• GPT (via OpenAI API)
• MiniMax
• Qwen
• Zhipu (GLM)
• Many others

Author's Choice: Codex OAuth

Reasoning:

• ChatGPT Pro membership holder
• Codex preferred for coding tasks
• Extremely high quota allocation
• OpenAI explicitly supports this usage
• Minimal risk compared to alternatives

CRITICAL: Claude Max Authorization Warning

⚠️ DO NOT use Claude Max credits for Clawdbot authorization

Why This Is Dangerous:

• Anthropic now restricts Max credits to Claude Code exclusively
• Using Clawdbot with Max credits may result in immediate account ban
• Multiple documented cases on X (Twitter) of account suspensions
• Anthropic taking aggressive enforcement action

Safe Alternatives:

• Claude API (separate billing)
• OpenAI models
• Chinese domestic models (MiniMax, Qwen, GLM)

Token Consumption Warning

Critical Issue: Clawdbot's context management is “incredibly poor”

Real-World Example:

• Team member testing with GLM-4.7
• Two days of testing
• Result: 30 million tokens consumed

Cost Implications:

• With Claude API pricing: potentially need to “sell your house” to afford 24/7 operation
• Single web scraping task (crawling dozen X articles): 1 million tokens consumed

Recommendation: Use cost-effective models or carefully monitor usage to avoid financial shock

Step 4: Channel Configuration

System asks which messaging platform to use:

Available Options:

• All international platforms listed
• Most Chinese users don't use these regularly

Recommendation: Select “Skip” (last option)

Reasoning: Tutorial will cover Feishu integration separately—more relevant for Chinese users

Step 5: Skills Installation

System asks about installing default skills:

Recommendation: Select “Yes” automatically

What Are Skills: Pre-built capabilities enabling specific tasks (document processing, data analysis, etc.)

Optional: Review each skill description and select individually, or skip entirely—skills can be added later through conversation

Step 6: Package Manager Selection

Choose installation method:

Common Options:

• npm (recommended—automatically available with Node.js)
• yarn
• pnpm

Author's Preference: npm for simplicity and universal availability

Step 7: Skills Marketplace

System presents available skills with descriptions:

Two Approaches:

Selective Installation: Review each skill and install relevant ones

Skip All: Proceed without skills—install later as needed through conversation with Clawdbot

Flexibility: Skills can be added/removed anytime, so initial choice isn't permanent

Step 8: Hooks Configuration

System asks about three plugins:

boot-md:

• Function: Automatically loads markdown text at startup
• Use Case: Inject rules, preferences, project background into every session
• Recommendation: Enable for consistent behavior

command-logger:

• Function: Logs executed commands and key operations
• Use Case: Problem diagnosis and session review
• Privacy Note: Disable if concerned about leaving traces
• Recommendation: Enable for troubleshooting capability

session-memory:

• Function: Saves conversation state and memory
• Use Case: Enables context continuity across sessions
• Recommendation: Enable for better experience

Author's Recommendation: Enable all three—all provide valuable functionality

Step 9: Installation Complete

Success Message: Installation finished successfully!

Startup Command:

clawdbot gateway --verbose

Default Port: 18789

Access URL: http://127.0.0.1:18789/chat

Result: Friendly lobster interface appears, ready for interaction

Feishu (Lark) Integration Tutorial

Why Feishu Integration Matters

Universal Desire: Everyone wants mobile/alternative chat app access—Clawdbot's greatest appeal

Chinese Market Solution: Feishu (Lark) currently easiest platform for Chinese users

Perfect Match: Feishu's bot capabilities align perfectly with Clawdbot's architecture

Community Plugin

GitHub Repository: https://github.com/m1heng/Clawdbot-feishu

Developer: @m1heng created Feishu connector for Clawdbot

The Manual Configuration Nightmare

Author's Experience:

• Followed GitHub instructions step-by-step
• Spent 1-2 hours debugging
• Constant errors and installation failures
• Frustration mounting

The Epiphany: “Wait, I'm an idiot. I have Clawdbot installed—it's supposed to be powerful. Why am I manually debugging? Let IT install itself!”

The Self-Installation Miracle

Simple Command in WebUI:

Install the Clawdbot plugin: clawdbot plugins install @m1heng-clawd/feishu

Result: Clawdbot successfully installed the plugin FOR ITSELF

Reaction: “What was that 1-2 hours of debugging even for??”

Lesson: Use the tool's capabilities instead of manual configuration when possible

Feishu Platform Configuration

Step 1: Visit Feishu Open Platform (open.feishu.cn)

Step 2: Create new application

Step 3: Record App ID and App Secret (critical for next steps)

The Magic Integration

Most Impressive Part: Send your App ID and App Secret directly in Clawdbot chat

What Happens: Clawdbot reads credentials and automatically completes remaining configuration

Experience: “Absolutely explosive satisfaction”

Feishu Platform Settings

Add Bot to Groups: Invite bot to desired Feishu groups

Enable Necessary Permissions (as specified in GitHub documentation):

• Message reading
• Message sending
• Group membership
• Event subscriptions
• Bot callbacks

Configure URLs and Endpoints (provided by Clawdbot after setup):

• Event subscription URL
• Card callback URL
• Message encryption settings

Final Steps

Publish Application: Submit for review and publishing in Feishu

Result: Chat with Clawdbot directly in Feishu!

Practical Testing: Invoice Processing Example

Real-World Scenario

Company Requirement: Expense reimbursement requires filling invoice data into Excel template

Traditional Approach: Manually entering each invoice line by line—tedious and time-consuming

Clawdbot Approach: Simple Feishu command initiates automatic processing

Execution Flow

User Action: Send instruction via Feishu

Clawdbot Operations:

1. Opens local files
2. Reads invoice content
3. Automatically fills Excel template
4. Saves completed document

Speed: Completed “in the blink of an eye”

Quality: “A bit rough, but has that flavor”

Remote Control Capability

Setup: Test computer left running continuously at office

Evening Use: Control from home using Feishu on phone

Experience: “Quite fun” to remotely command office computer

The Philosophical Dilemma

Why Clawdbot Exploded

“Clawdbot's explosion reflects people's extreme desire for AI to actually get work done.”

Current Attitude: Author feels conflicted about recommendation

The Security Balance Sheet

Capability: Can help complete tasks efficiently

Risk: May modify or delete things during inattention

Detection: You might not even notice until much later

Emotional Response: When discovery happens, “blood rushes to your head” with intense frustration

The Cautious Recommendation

Warning: Definitely proceed cautiously

Curiosity: Exploration of unknown territory is genuinely interesting

Target Audience: Those with curiosity about new technology and spare computer for experimentation

Value Proposition: Ultimate proactive local agent experience will “refresh your imagination boundaries about AI”

The Future Vision

Current Perception: May seem “a bit offensive”

Future Reality: Isn't this what we envision for ideal agents?

Core Difference: Only accuracy and error tolerance rates separate current from ideal

The Magic Moment: Lying in bed, remembering something, opening Feishu/QQ (potentially WeChat in future), sending message—then Clawdbot diligently works on your behalf

Experience: “Really addictive and quite magical”

The Key Management Metaphor

Before Handing Over Keys

Critical Question: Which doors should these keys open?

Household Staff Analogy:

Common Spaces (Kitchen, Living Room, Closet):

• Probably comfortable granting access
• Low risk for these areas

Private Spaces (Bedroom, Personal Office):

• Would you confidently give keys to household help?
• Uncertain at best

AI Usage Parallel

Clawdbot Reality: Genuinely interesting and addictive

Essential Requirement: Must think clearly about what access to grant

Core Consideration: This reveals your actual management and delegation abilities

The Universal Principle

After Careful Thought: Maintain eternal curiosity about the world

Balance: Responsible deployment with adventurous exploration

Conclusion: Deploy Responsibly, Explore Fearlessly

Final Assessment

Clawdbot represents a significant milestone in AI evolution—from passive tools to active assistants. The explosive growth and passionate community engagement demonstrate genuine desire for AI that executes rather than merely advises.

Three-Point Action Plan

1. Understand Risks Completely:

• High system permissions enable powerful capabilities
• Same permissions create severe vulnerability
• One mistake could have lasting consequences

2. Deploy Safely:

• Use spare device, never primary computer
• Consider virtual machine or cloud deployment
• Start with limited permissions, expand gradually

3. Explore Thoughtfully:

• Test thoroughly before critical applications
• Monitor behavior and token consumption
• Maintain human oversight of important operations

The Bigger Picture

Clawdbot foreshadows a future where AI assistants proactively manage tasks, remember context across sessions, and integrate seamlessly into daily workflows. Current limitations in accuracy and error handling will improve, but the fundamental paradigm shift has begun.

Who Should Try Clawdbot?

Ideal Candidates:

• Technology enthusiasts with spare hardware
• Developers wanting to experiment with agentic AI
• Early adopters comfortable with bleeding-edge tools
• Organizations with proper security infrastructure

Who Should Wait:

• Those relying on primary computer for critical work
• Users without technical support resources
• Anyone uncomfortable with high permission requirements
• Situations requiring guaranteed reliability

The Permission Philosophy

Think of Clawdbot like hiring a household manager:

Grant Access To:

• Non-sensitive routine tasks
• Automatable repetitive work
• Well-defined workflows with clear boundaries

Restrict Access From:

• Personal documents and private files
• Financial accounts and credentials
• Irreplaceable data without backups
• Mission-critical systems

Final Words

If you've read this far and found value, please:

• Like and share this article
• Star the project for updates
• Deploy responsibly with full awareness
• Maintain curiosity about AI's future

Remember: The most advanced AI assistant is only as safe and effective as the human judgment guiding it.