How to Set Up Clawdbot/Moltbot: Your Self-Hosted AI That Actually Does Things

By Lucia Clifford, CTO and Founder of Sharpei.ai and iancloud.ai

What is Clawdbot/Moltbot?

Clawdbot is an open-source AI assistant that takes a different approach: it proactively messages you instead of waiting for prompts.

Unlike cloud-based chatbots that just answer questions, Clawdbot actually does things — it runs commands, automates workflows, manages files, controls your browser, and reaches out to you with morning briefings, calendar alerts, and reminders on its own.

And because it runs locally, your data stays yours. Conversations and memories are stored as Markdown files on your machine. No third-party servers. No data leaks. MIT licensed. Completely free.

Think of it as a smart assistant that lives on your computer (or server), proactively helping you via WhatsApp, Telegram, Slack, or any messaging app you already use.

Why Clawdbot Stands Out

Most AI tools are conversational. You ask, they answer, you copy-paste.

Clawdbot is operational. It doesn’t just tell you how to do something — it does it.

Key Features

1. Runs Locally — Your Data Stays With You (Or not)

• Runs on Mac, Windows, Linux, or any VPS/cloud server

• All memory, files, and logs stored on your machine or your team’s server

• Run it fully local for privacy or on EC2 for team access

• Your choice: air-gapped personal assistant or shared team infrastructure

⚠️ Let's be honest about "local": Self-hosted doesn't mean your data stays entirely private. Unless you're running local models via Ollama, your prompts and context go to AI providers (Anthropic, OpenAI, etc.) And there have been claims of vulnerabilities exposing sensitive data in misconfigured instances, hundreds of Clawdbot setups have been found open on the internet.

The advantage isn't no one sees your data, it's that you control where it lives and who can access the gateway. That's still a big deal compared to SaaS tools, but don't confuse self-hosted with air-gapped.

2. Multi-Platform Messaging

Talk to Clawdbot through your favorite apps:

Message your AI from your phone, your desktop, anywhere.

3. Real Task Automation

Clawdbot doesn’t just answer questions, it executes:

**Terminal commands** — run scripts, manage servers

**File operations** — organize, rename, process documents

Browser automation — scrape data, fill forms, take screenshots

Data analysis — process Excel files, generate reports

Email & calendar — schedule, send, organize

API integrations — connect to any service

Smart home — control devices, routines

If it’s on your computer or accessible via API, Clawdbot can handle it.

4. Persistent Memory (Stored as Markdown)

Clawdbot remembers past interactions, with conversations and memories stored as Markdown files on your machine. These files are human-readable, version-controllable, and portable meaning the AI gets smarter and more personalized over time, with context carrying across sessions.

5. Extensible Skills & Plugins

Built-in integrations for Notion, GitHub, Trello, Slack, weather, audio transcription (Whisper), and more. Add custom skills for your specific workflows. Easy to configure from their installation wizard.

6. Flexible AI Models

Works with:

Claude(Anthropic) — best for complex reasoning

GPT-4(OpenAI) — widely supported

Local models via Ollama— completely private, no API costs

Any AI API — bring your own

Run entirely offline with Ollama, or use cloud models. You choose the brain.

7. Proactive Alerts & Automation

Unlike chatbots that wait for you, Clawdbot reaches out:

• Morning briefings (weather, calendar, priorities)

• Scheduled reminders

• Triggered alerts (Sentry errors, new tickets, etc.)

SETUP GUIDE — Let´s cut to the fun stuff 🎉

Yep thats me 😜

Why I Don’t Run This Locally

First lesson learned: running Clawdbot on my MacBook was killing my RAM. The gateway, the agents, the WhatsApp bridge, all competing with my browser tabs and Docker containers. My fans sounded like a jet engine.

So I moved everything to the cloud. Cheap EC2 instance, always on, never slowing down my actual work machine. My laptop just connects to it.

In my next article, I’ll show you how I use this setup to replace junior engineers 🤯, a full automation pipeline that turns Trello cards into merged PRs while I sleep. But first, let’s get you set up.

0. Spin Up an EC2 Instance (Skip If Running Locally)

If you want to run Clawdbot on your laptop, skip this. But if you want it always-on without murdering your RAM, you need a cloud server.

Here’s the fun part: I didn’t even log into AWS.

I just told https://iancloud.ai: “spin up an EC2 for Clawdbot” and it did. Created the instance, configured the security group, installed Node.js, set up Cloudflare Tunnel. All I did was approve the commands.

If you want to do it manually:

• AMI: Ubuntu 22.04 LTS

• Instance type: t3.small (~$15/month) — t3.micro chokes

• Security group: SSH only (port 22). Cloudflare Tunnel handles the rest.

1. Install Clawdbot

npm install -g clawdbot
clawdbot setup

This installs the CLI and walks you through initial configuration: API keys, default model, etc.

2. Start the Gateway

clawdbot gateway install   # Auto-starts on boot

The gateway is the brain. It receives messages from your channels (WhatsApp, Slack, etc.), routes them to the right agent, and sends responses back. Runs on port 18789 by default.

3. Set Up Remote Access (Cloudflare Tunnel)

brew install cloudflared   # or apt install on Linux
cloudflared tunnel login
cloudflared tunnel create clawdbot
cloudflared tunnel --url http://localhost:18789

❓ Why Cloudflare Tunnel? No exposed ports, no static IP needed, free TLS encryption, works behind NAT. Your EC2 (or Mac) becomes securely accessible from anywhere without the security nightmare of opening ports.

Make it persistent: The quick tunnel above is temporary. For a permanent URL:

# On EC2 — run as a service
cloudflared service install
cloudflared tunnel route dns clawdbot your-subdomain.yourdomain.com

3b. Connect Your Laptop to the Remote Gateway (Optional)

If you’re running Clawdbot on EC2 but want to interact from your laptop:

# On your laptop
clawdbot node install \
 - host your-tunnel-url.trycloudflare.com \
 - port 443 \
 - tls

Now your local machine connects to the remote gateway. You can run clawdbot commands locally and they execute on EC2.

4. Connect WhatsApp

clawdbot channels login # Scan QR with WhatsApp

Scan the QR code with your phone. Done!

Your AI now has a phone number.

⚠️ Critical: Configure an allowlist so random people can’t message your AI:

{"channels": {"whatsapp": {"dmPolicy": "allowlist", "allowFrom": ["+1234567890"]}}}

Save this in ~/.clawdbot/clawdbot.json.

5. Add Your Integrations

# Notion - for reading your docs and knowledge base
echo "ntn_your_key" > ~/.config/notion/api_key
# Trello - for task automation (more on this in my next article)
echo "your-api-key" > ~/.config/trello/api_key
echo "your-token" > ~/.config/trello/token
# GitHub - for creating PRs and managing repos
gh auth login

6. Create Agents

clawdbot agents add - workspace ~/Projects work
clawdbot agents add - workspace ~/Documents personal

Agents are isolated workspaces. Each one has its own context, memory, and file access. I have separate agents for each codebase — keeps things clean.

7. Schedule Automations

clawdbot cron add - name "morning-briefing" - cron "0 8 * * *" \
 - message "Weather, calendar, top emails" - deliver - to "+1234567890"

This is where Clawdbot becomes proactive. Set up cron jobs to have it reach out to you — morning briefings, end-of-day summaries, or (like I do) automated spec writing and PR creation.

Why I Love This

🏔️ Snowboarding → beer → check WhatsApp → ”yep, code it” → 💥 PR created → review after my shower.

The life I deserve 🩵

At Sharpei, we use Clawdbot to automate dev workflows.

At iancloud.ai, we’re making this setup a one-click.