How to Set Up Clawdbot: Personal AI Assistant Guide

If you haven’t heard of Clawdbot yet, you will soon. It went from a personal side project by Peter Steinberger (founder of PSPDFKit) to 30,000+ GitHub stars in January 2026 alone, a rise we covered in more detail in the Clawdbot/Moltbot/OpenClaw viral saga. After spending a couple of weeks using it as my daily AI assistant, I understand why.

Clawdbot is an open-source, self-hosted AI assistant that lives inside the messaging apps you already use: WhatsApp, Telegram, Discord, Slack, Signal, even iMessage. Unlike ChatGPT or Claude in a browser tab, Clawdbot is always on, remembers your entire conversation history, and can actually do things on your computer. It can browse the web, manage your email, run terminal commands, and automate workflows.

This guide walks you through setting it up from scratch, whether you’re on a Mac, Linux box, or Windows machine.

Clawdbot connects your messaging apps to a powerful local AI agent

What Makes Clawdbot Different

Clawdbot differs from ChatGPT or Claude in a browser because it has persistent memory across weeks, lives inside apps you already use, and can take real actions like browsing the web or running terminal commands, not just answering questions in a session-based chat window.

Feature	ChatGPT/Claude	Clawdbot
Memory	Session-based, limited	Unlimited, persistent across weeks/months
Where it lives	Browser tab	Your messaging apps (WhatsApp, Telegram, etc.)
Can take actions	Limited	Full computer access (browser, terminal, files, email)
Proactive	Never	Can send you alerts, briefings, reminders
Data ownership	Cloud (OpenAI/Anthropic servers)	Fully local, you control everything
Cross-device	Separate sessions	Unified (start on phone, continue on desktop)
Cost	$20/month subscription	Pay-per-use API ($20-50/month typical)
Customizable	System prompts only	Full personality, skills, and workflow customization

The persistent memory is the killer feature for me. I can reference a conversation from two weeks ago and it knows exactly what I’m talking about. No “I don’t have access to previous conversations” nonsense.

Prerequisites

Before you start, make sure you have:

• Node.js 22 or newer. Clawdbot requires Node 22+
• An API key. Anthropic (Claude) recommended for best results, or OpenAI
• A messaging app. WhatsApp, Telegram, Discord, Slack, or others
• Terminal access. macOS Terminal, Linux shell, or Windows PowerShell

Which AI Model Should You Use?

Provider	Model	Cost (per 1M tokens)	Best For
Anthropic	Claude Sonnet	$3 input / $15 output	Best balance of quality and cost
Anthropic	Claude Opus 4.5	$15 input / $75 output	Complex tasks, long context
OpenAI	GPT-4o	$2.50 input / $10 output	General-purpose, good value
OpenAI	GPT-4o-mini	$0.15 input / $0.60 output	Budget-friendly, lighter tasks
Local	Ollama models	Free (hardware cost)	Privacy-first, no API dependency

I’d recommend starting with Claude Sonnet via Anthropic API. It’s the sweet spot for agentic tasks: smart enough to handle complex workflows, affordable enough for daily use. Typical usage runs $20-50/month. If you’d rather avoid API costs entirely, our guide to running local LLMs with Ollama covers what to expect from the local-model route.

Installation

macOS & Linux

Open your terminal and run:

curl -fsSL https://clawd.bot/install.sh | bash

That’s it. The script installs Node.js if needed and sets up the Clawdbot CLI.

Windows

Open PowerShell as Administrator and run:

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

Important: WSL2 (Windows Subsystem for Linux) is strongly recommended over native Windows. If you’re comfortable with WSL, install Ubuntu via the Microsoft Store, then follow the macOS/Linux steps inside WSL instead. You’ll have a better experience.

From Source (For Developers)

If you want to contribute or run the latest development version:

git clone https://github.com/clawdbot/clawdbot.git
cd clawdbot
pnpm install
pnpm ui:build
pnpm build
clawdbot onboard --install-daemon

For active development with hot reload:

pnpm gateway:watch

The Onboarding Wizard

This is where the magic happens. Run:

clawdbot onboard --install-daemon

The --install-daemon flag is important. It sets up Clawdbot as a background service so it starts automatically and stays running. The wizard walks you through everything:

Step 1: Risk Acknowledgment

Clawdbot will ask you to confirm that you understand what it can do. This isn’t just a formality. You’re giving an AI assistant access to your browser, files, and terminal. Take this seriously.

Step 2: Choose Setup Mode

You’ll see options like Quickstart and Advanced. If this is your first time, go with Quickstart. You can always reconfigure later.

Step 3: Select Your AI Provider

Pick your LLM provider:

• Anthropic (recommended): paste your API key from console.anthropic.com
• OpenAI: paste your API key from platform.openai.com/api-keys

If you’re already getting Claude-pilled with Claude Code, you can also reuse those credentials with claude setup-token.

Step 4: Choose Your Model

Select which specific model to use. For Anthropic, Claude Sonnet is the recommended default. You can upgrade to Opus 4.5 for heavier tasks later.

Step 5: Connect a Messaging Channel

This is where you pick where Clawdbot will live. Your options include:

• WhatsApp: scan QR code via Linked Devices
• Telegram: create a bot via @BotFather
• Discord: create a bot application
• Slack: install as a workspace app
• Signal, iMessage, Google Chat, Microsoft Teams, Matrix: all supported

You can skip this step and just use the built-in web dashboard to start chatting immediately.

Step 6: Configure Skills

Skills are modular plugins that extend what Clawdbot can do. The wizard lets you enable:

• Web search: uses Brave Search API (2,000 free queries/month)
• Browser control: navigate websites, fill forms, scrape data, similar in spirit to the AI browser agents now shipping in Comet, Atlas, and Mariner
• Email: Gmail integration for reading and sending
• Calendar: Google Calendar management
• GitHub: PR reviews, issue management, code operations

Enable what you need now. You can add more later.

Step 7: Done

The wizard installs the Gateway daemon and you’re ready to go.

Verifying Your Installation

After setup, run these commands to make sure everything is working:

# Check service status
clawdbot gateway status

# Run health check
clawdbot health

# Security audit (recommended)
clawdbot security audit --deep

If the gateway isn’t running, start it manually:

clawdbot gateway --port 18789 --verbose

The Dashboard

The fastest way to start chatting, no channel setup needed:

clawdbot dashboard

This opens your browser at http://127.0.0.1:18789/ with a web chat interface. You can start talking to your AI assistant immediately.

Setting Up Messaging Channels

WhatsApp

This is the most popular channel and the one I use daily.

clawdbot channels login

A QR code appears in your terminal. On your phone, open WhatsApp, go to Settings > Linked Devices > Link a Device, and scan the code.

Once linked, message yourself (or the bot number) and you’ll get a pairing code. Approve it:

clawdbot pairing approve whatsapp <code>

Now you can message your AI assistant on WhatsApp just like texting a friend.

Telegram

1. Open Telegram and message @BotFather
2. Send /newbot and follow the prompts to create a bot
3. Copy the bot token
4. Paste it into Clawdbot’s configuration when prompted

Discord

1. Go to the Discord Developer Portal
2. Create a new application
3. Enable Message Content Intent under Bot settings
4. Copy the bot token
5. Add the bot to your server
6. Configure in Clawdbot

Other Channels

Slack, Signal, iMessage, Google Chat, Microsoft Teams, and Matrix are all supported. Each has its own setup flow. The onboarding wizard or clawdbot channels login will guide you through it.

Pairing & Security

This is important and often overlooked.

By default, Clawdbot uses a pairing system. When an unknown person messages your bot, it generates a pairing code instead of responding. You must explicitly approve each person:

# See pending pairing requests
clawdbot pairing list whatsapp

# Approve a specific request
clawdbot pairing approve whatsapp <code>

This prevents random people from using your AI assistant (and running up your API bill). It’s also a meaningful defense against the kind of AI agent security incidents that happen when an agent with terminal and browser access takes instructions from the wrong source.

Security Best Practices

• Run clawdbot security audit regularly
• Use a dedicated phone number for WhatsApp, not your personal one
• Never add the bot to public group chats. Treat it as a private terminal
• Scope API tokens tightly (minimum permissions for Gmail, GitHub, etc.)
• Enable sandbox mode for risky operations
• Whitelist specific terminal commands instead of allowing unrestricted access

Customizing Your Assistant

The SOUL.md File

This is the personality file that defines how your assistant behaves. It lives at ~/clawd/SOUL.md and you can edit it freely.

Here’s what a basic SOUL.md might look like:

# Personality

You are my personal AI assistant. Be direct, concise, and practical.
Don't be overly formal. We're on WhatsApp, not in a boardroom.

# Context

- I'm a software developer based in [your city]
- I work primarily with TypeScript, React, and Node.js
- My work hours are 9am-6pm, don't send proactive messages outside those
- I prefer bullet points over paragraphs

# Rules

- Always confirm before sending emails on my behalf
- Never execute destructive commands (rm -rf, drop database, etc.) without asking
- When researching, cite your sources
- If you're unsure about something, say so

The more context you give it, the more useful it becomes. I’ve gradually built mine up over weeks, and the assistant gets better as it learns my preferences.

Adding Skills

Skills extend Clawdbot’s capabilities. They live in ~/clawd/skills/ as SKILL.md files.

You can browse and install skills from ClawdHub, the community skill registry. It plays a similar role to the tool registries covered in our MCP servers practical guide, just scoped to Clawdbot’s own skill format:

clawdbot skills list
clawdbot skills install <skill-name>

Or create your own by adding a SKILL.md file in a subfolder of ~/clawd/skills/.

Chat Commands

Once you’re chatting with Clawdbot (via any channel), these commands are available:

Command	What It Does
/status	Show session info: model, tokens used, cost
/new or /reset	Start a fresh session
/compact	Compress context (summarize history to save tokens)
/think <level>	Set thinking depth: off, minimal, low, medium, high, xhigh
/verbose on/off	Toggle detailed output
/usage off/tokens/full	Control usage display
/restart	Restart the gateway

The /compact command is particularly useful when conversations get long. It summarizes the history to reduce token usage without losing important context.

Architecture Overview

Understanding the architecture helps with troubleshooting:

Your Phone (WhatsApp/Telegram/Discord/etc.)
         ↓
┌─────────────────────────────────┐
│   Gateway (control plane)       │
│   ws://127.0.0.1:18789          │
│                                 │
│   ├── Agent (Claude/GPT brain)  │
│   ├── Skills (web, email, etc.) │
│   ├── Memory (local storage)    │
│   └── Tools (browser, terminal) │
└─────────────────────────────────┘
         ↓
  Your Computer (browser, files, apps)

The Gateway is the central hub. It runs locally on your machine, connects to your messaging channels, routes messages to the AI agent, and executes tool calls. Everything stays on your hardware: messages, memory, and actions.

Remote Access with Tailscale

If you’re running Clawdbot on a Mac Mini or home server and want to access it from anywhere:

# Tailscale serve (accessible only within your tailnet)
# Configure in gateway settings:
# gateway.tailscale.mode: serve

# Tailscale funnel (accessible from public internet)
# gateway.tailscale.mode: funnel
# Requires password auth: gateway.auth.mode: "password"

SSH tunnels also work if you prefer that approach.

Cost Management

API costs can add up if you’re not careful. Here’s what to expect:

Usage Level	Monthly Cost	Description
Light	$10-20	A few messages per day, simple tasks
Moderate	$20-50	Daily use, research, email management
Heavy	$50-150	Complex automations, multi-step workflows
Power user	$150-300+	Multiple agents, continuous automation

Tips to Keep Costs Down

• Use Claude Sonnet instead of Opus for everyday tasks
• Use /compact regularly to reduce context size
• Set thinking level to low or minimal for simple queries
• Use gpt-4o-mini for simple tasks if on OpenAI
• Consider running a local model via Ollama for basic queries, with cloud fallback for complex tasks

Troubleshooting

Gateway Won’t Start

# Check if something else is using port 18789
lsof -i :18789

# Start with verbose logging
clawdbot gateway --port 18789 --verbose

# Check Node version (needs 22+)
node --version

WhatsApp Connection Drops

This is the most common issue. WhatsApp Web sessions expire periodically.

# Re-login
clawdbot channels login

Scan the QR code again. Your conversation history is preserved.

High API Costs

# Check usage
clawdbot status

Review which conversations are consuming the most tokens. Use /compact in long-running sessions.

Messages Not Being Delivered

Check pairing status:

clawdbot pairing list <channel>

If messages from you aren’t being processed, you may need to re-approve pairing.

Run the Doctor

When in doubt:

clawdbot doctor

This checks for common configuration issues and risky settings.

Real-World Use Cases I’ve Found Useful

After two weeks of daily use, here’s what I actually use Clawdbot for:

Morning briefing. I set up a cron job that sends me a summary of my calendar, unread important emails, and weather at 7:30am every morning via WhatsApp. I read it while having coffee.

Research on the go. When I’m reading something on my phone and want to dig deeper, I just message Clawdbot. It searches the web, summarizes findings, and sends them back, all without opening a browser.

Quick code help. I snap a screenshot of an error, send it to Clawdbot on WhatsApp, and get a fix. The multimodal understanding works surprisingly well.

Email triage. “Check my email and flag anything urgent.” It scans my inbox, categorizes messages, and gives me a summary. I respond to the important ones and ignore the rest.

Meeting prep. “I have a meeting with [company] tomorrow, give me a briefing.” It researches the company, checks our CRM for history, and sends me talking points.

Who Should (and Shouldn’t) Use Clawdbot

Good fit if you:

• Want a genuinely useful AI assistant, not just a chatbot
• Are comfortable with terminal/command line
• Care about data ownership and privacy
• Want persistent memory across conversations
• Use messaging apps as your primary communication tool

Not a good fit if you:

• Want zero-setup, just-works simplicity
• Aren’t comfortable giving AI access to your system
• Need enterprise-grade reliability guarantees
• Don’t want to manage API costs

What’s Next

Clawdbot is moving fast. The community is active (5,000+ Discord members), new skills are being published regularly on ClawdHub, and the development pace is impressive. Companion apps for macOS, iOS, and Android add native integration with cameras, screen recording, and voice, turning your phone into a true AI-powered assistant.

If you’ve been waiting for AI assistants to move beyond browser tabs and actually integrate into your life, Clawdbot is the closest thing I’ve found to that vision.

---

Resources

• Official Documentation
• GitHub Repository
• Discord Community
• ClawdHub Skills Registry
• Beebom: Setup on Mac Mini
• Beebom: Setup on Windows

Want help setting up Clawdbot for your team or integrating it into your workflow? Talk to CODERCOPS. We’ve deployed personal AI assistants for individuals and teams alike.