How to Use Moltbot: Beginner's Guide to Your Self-Hosted AI Assistant

What's up, self-hosting enthusiasts — if you've ever looked at a "personal AI assistant" and thought "yeah, but can I actually run this on my own hardware without it phoning home?" then you're in the right place.

I've spent the last few weeks running Moltbot (formerly Clawdbot) through actual daily tasks. Not the polished setup videos you see on YouTube. Real work: file sorting, automated reminders, GitHub integrations, the stuff that either works or breaks your flow.

Here's what I found: Moltbot isn't your typical chatbot. It's a self-hosted personal AI assistant that actually lives on your machine — a Mac Mini, VPS, or even a Raspberry Pi. According to the official Moltbot documentation, it connects to messaging platforms you already use (WhatsApp, Telegram, Slack, Discord) and can execute real tasks on your behalf.

The question I kept asking myself: can this replace the friction of switching between five different apps just to get one thing done?

Let me show you what actually happened when I put it through real use.

Getting Started

The first thing you need to know: Moltbot runs locally. That means you're not sending your data to some cloud service. Developed by Austrian engineer Peter Steinberger, Moltbot is designed to run on your own hardware and maintain persistent context.

Here's the reality check — setup isn't instant. But if you've ever configured a development environment, you can handle this.

What you'll need:

Quick start via CLI wizard:

# Install (one-line installer)
curl -fsSL https://molt.bot/install.sh | bash
exec bash

# Run onboarding wizard
moltbot onboard --install-daemon

The onboarding wizard walks through gateway setup, workspace configuration, channel linking, and skills installation. It took me about 15 minutes the first time I ran it.

The wizard will ask you:

• Which AI model to use (I went with Claude Sonnet — more on that later)
• Which messaging apps to connect
• Security settings (don't skip this)

Once it completes, the gateway runs as a background service. That means Moltbot stays alive even when you're not actively using it.

First message test:

Open WhatsApp (or your chosen platform), find the chat with your linked number, and send:

Hello!

If Moltbot responds, you're in. If it doesn't, check:

• Gateway is running: moltbot status --all
• Authentication is configured: moltbot health

Basic Commands

Sending Messages

Here's where I hit my first confusion point. Moltbot doesn't work like ChatGPT where you just type and wait. It's proactive.

Natural language works:

Good morning. Any important emails today?

Moltbot can answer with email summaries and can proactively send morning briefings or reminders when certain events occur.

But so do explicit commands:

# Via CLI (useful for automation)
moltbot message send --to +1234567890 --message "Deploy complete"

# Via agent (for AI processing)
moltbot agent --message "Ship checklist" --thinking high

I use CLI commands for scripts and cron jobs. Natural language for everything else.

What I learned the hard way: Moltbot maintains context across platforms. If you start a conversation on WhatsApp, you can continue it via Telegram or the web dashboard — it remembers. The system maintains a single conversation view across all platforms, preserving context regardless of which channel you're using.

Using Skills

Skills are Moltbot's superpower. Think of them as modular capabilities that extend what the AI can do.

Moltbot has over 130 contributors and features an extensible design that allows unlimited customization via skills and plugins.

Pre-installed skills:

Out of the box, Moltbot comes with 50+ bundled skills:

• Weather checks
• GitHub operations
• Notion integration
• Slack automation
• File system operations
• Browser control

Installing new skills:

# Search for what you need
moltbot skill search "calendar"

# Install a skill
moltbot skill install google-calendar

Real example from my workflow:

I needed to automate GitHub issue tracking. Here's what I did:

Hey Moltbot, check if there are any new issues in the project repo

The GitHub skill kicked in, fetched the issues, and gave me a summary. No opening browser. No switching apps.

Important note: Some skills require external tools to be installed separately — Moltbot will notify you if dependencies are missing when you try to use them.

Setting Reminders

This is where Moltbot starts feeling like JARVIS.

Simple reminder:

Remind me to review the deployment at 3 PM

Recurring reminder:

Every Monday at 9 AM, send me the weekly checklist

Moltbot supports cron integration and wakeups, enabling scheduled and conditional automation.

The game-changer: Moltbot can proactively reach out. Unlike traditional assistants that wait for your command, it can message you first when conditions are met.

I set it up to notify me when:

• Disk usage exceeds 80%
• A specific GitHub label is added to issues
• Someone mentions me in Slack

Monitor disk space and alert me if it goes above 80%

It worked. Two weeks in, I got a message at 2 AM (yes, it works 24/7) warning me about storage. Saved me from a potential crash.

Daily Workflows

Let me show you how this actually plays out in real work.

Morning routine automation:

Instead of manually checking five different places, I ask:

Morning brief

Moltbot responds with:

• Unread emails summary
• Calendar for the day
• Any overnight notifications
• Weather forecast

File organization workflow:

I have a messy Downloads folder. Instead of sorting manually:

Organize my downloads folder — move PDFs to Documents, images to Photos

Moltbot executes shell commands directly, bypassing the GUI to perform file operations like moving, organizing, and managing directories.

It runs in the background via shell commands:

mv ~/Downloads/*.pdf ~/Documents/
mv ~/Downloads/*.{jpg,png} ~/Photos/

Results: Downloads organized in seconds.

Code deployment workflow:

When I'm ready to deploy:

Run the deployment checklist:
1. Run tests
2. Build production
3. Deploy to staging
4. Notify team on Slack

Moltbot can execute each step, report progress, and handle failures. It's not magic — it's using skills like shell access, Slack integration, and GitHub webhooks — but the experience feels seamless.

Meeting notes workflow:

During meetings, I send quick voice notes via WhatsApp. Moltbot supports voice input through messaging apps, automatically transcribing voice notes sent via WhatsApp or Telegram.

Later, I ask:

Summarize the meeting notes from today and create action items

It pulls from memory, structures the notes, and lists actionable next steps.

Tips and Tricks

Here's what I wish I knew from day one:

1. Use the web dashboard for setup, messaging for speed

Access the dashboard:

moltbot dashboard
# Opens http://127.0.0.1:18789/

The Control UI is great for:

• Viewing full conversation history
• Adjusting settings
• Checking system health

But for quick interactions, messaging platforms are faster.

2. Customize the personality

Moltbot's behavior is defined by Markdown files in your workspace:

~/.clawdbot/agents/main/workspace/
├── IDENTITY.md      # Who Moltbot is
├── SOUL.md          # Personality traits
├── USER.md          # What it knows about you
├── TOOLS.md         # Available capabilities
└── memory/          # Long-term memories

I edited USER.md to include:

• My time zone
• Preferred communication style
• Project context
• Common tasks

This made responses more relevant and reduced back-and-forth clarifications.

3. Leverage persistent memory

Unlike stateless AI, Moltbot maintains persistent memory through Markdown files that store long-term context, preferences, and conversation history.

Every interaction builds context. The longer you use it, the better it understands your patterns.

4. Set up SSH tunnel for remote access

If you're running Moltbot on a VPS but want to access the dashboard from anywhere:

# On your local machine
ssh -L 18789:localhost:18789 user@your-vps-ip

# Access dashboard at
http://localhost:18789

5. Monitor resource usage

Moltbot runs continuously. Keep an eye on:

moltbot status --all  # Full system status
moltbot health        # Quick health check

6. Use local models to reduce costs

Don't want to pay for API calls? You can run Moltbot with local models like GLM-4.7-Flash for tool-calling capabilities.

Configure in settings or via wizard:

moltbot configure --section models

7. Sandbox for safety

Moltbot acts as a bridge between language models and your local operating system, executing commands with full system access.

This is powerful but risky. Enable sandboxing for non-critical sessions:

agents:
  defaults:
    sandbox:
      mode: "non-main"

This isolates group/channel sessions from your main agent that runs on host.

8. Start small, expand gradually

I didn't enable all 50 skills at once. I started with:

• File operations
• GitHub
• Slack
• Reminders

Then added more as I identified friction points in my workflow.

Security reality check:

Let's talk about the elephant in the room. Security researchers identified vulnerabilities in publicly accessible Moltbot instances, including risks of credential theft and remote code execution.

Best practices:

• Never expose Moltbot directly to the internet
• Use Tailscale or SSH tunnels for remote access
• Set up proper authentication (the wizard guides you)
• Allowlist specific contacts in messaging platforms
• Review the official security guide

Common mistakes I made (so you don't have to):

1. Forgetting to set authentication — Gateway won't respond without it
2. Not backing up the workspace — Lost all context after reinstalling
3. Enabling too many channels at once — Created message routing chaos
4. Ignoring system resources — Ran out of RAM on a 1GB VPS

The Bottom Line

After three weeks of daily use, here's my take:

Moltbot delivers on the promise IF:

• You're comfortable with command-line setup
• You want full control over your data
• You can tolerate occasional rough edges
• You're willing to invest time in configuration

It's not for you if:

• You need plug-and-play simplicity
• You want a polished commercial product
• You're not technical

Where it genuinely impressed me:

• Persistent memory that actually works
• Proactive automation (it messages me first)
• Cross-platform context (WhatsApp to Telegram seamlessly)
• Local-first privacy

Where it's still rough:

• Setup complexity (getting better with each update)
• Documentation scattered across GitHub, Discord, and docs site
• Some skills require manual tool installation
• Security requires careful configuration

The comparison to JARVIS isn't hype. Moltbot became one of the fastest-growing open-source projects, reaching over 60,000 GitHub stars by offering a "24/7 Jarvis" experience where the AI can proactively execute autonomous tasks.

But remember: it's self-hosted infrastructure you manage, not a turnkey SaaS product.

Is it worth trying?

If you're the kind of person who runs their own servers, tinkers with automation, and values data sovereignty — absolutely. You'll find real productivity gains once you get past the setup curve.

If you want something that "just works" without configuration — look elsewhere. Or wait for the ecosystem to mature.

For me, the ability to automate my entire morning routine with one message, have my AI assistant remember project context across weeks, and execute shell commands on my behalf makes it worth the initial investment.

Your mileage will vary based on your technical comfort level and specific use cases.

The setup friction you just read about? That's exactly what we eliminated in Macaron. Persistent memory, task automation, and custom tool creation—minus the gateway configuration, skill installation, or security audit checklist. Try one of your daily tasks and judge the results yourself. Free to start, no VPS hosting, reversible anytime.