Build a $100 AI Agent Server: Run OpenClaw on Raspberry Pi 24/7

Imagine having a personal AI assistant that’s always available, completely private, and costs less than a dollar a month to run. Sound too good to be true? With OpenClaw and a Raspberry Pi, it’s entirely possible—and I’m going to show you exactly how to build one.

Why Run an AI Agent on Raspberry Pi?

The concept is simple but powerful: turn a $100 piece of hardware into a 24/7 AI-powered assistant that lives in your home. No cloud subscriptions eating your data, no monthly fees, no privacy concerns about your conversations being stored on someone else’s servers.

Here’s what makes this setup compelling:

• 24/7 Availability: Your AI assistant never sleeps. It’s ready to help at 3 AM or 3 PM.
• Ultra-Low Power: The Raspberry Pi draws about 5 watts—roughly the same as a phone charger. That translates to $0.43-$1.01 per month in electricity costs.
• Complete Privacy: Every conversation, every task, every piece of data stays on your network. Nothing leaves your home.
• Compact & Quiet: The Pi is credit-card sized and can run silently with passive cooling. Tuck it in a drawer and forget it’s there.

What is OpenClaw?

OpenClaw is an open-source AI agent framework that’s been making waves in the developer community. Originally created as a weekend project by Peter Steinberger, it has grown to over 68,000 GitHub stars—and for good reason.

Unlike a simple chatbot, OpenClaw gives AI models real agency. It can:

• Execute shell commands and write code
• Connect to Telegram, Discord, Slack, Signal, and WhatsApp
• Integrate with tools like GitHub, Home Assistant, Notion, and Google Calendar
• Run scheduled tasks (cron jobs) for automated workflows
• Maintain persistent memory across conversations

Think of it as giving an LLM a computer to control. The AI doesn’t just answer questions—it actually does things.

Hardware You’ll Need

Let’s talk about the physical components. You don’t need much, but choosing the right hardware makes the difference between a smooth experience and constant frustration.

The Raspberry Pi Itself

Recommended: Raspberry Pi 5 (8GB)

The 8GB model is the sweet spot for running OpenClaw. At around $80, it provides enough RAM for:

• Loading AI models with decent context windows
• Running multiple agents simultaneously
• Handling background skills without memory pressure

Budget Option: Raspberry Pi 4 (8GB or 4GB)

The Pi 4 8GB works great for most use cases. The 4GB model is viable for single-user, basic setups but you’ll feel the limits with heavier workloads.

Storage: Skip the SD Card (Eventually)

Here’s a tip many tutorials miss: SD cards die. The constant writes from logs, memory files, and transcripts will corrupt your card within months.

Best Options:

1. USB SSD ($25-40): Plug in a small USB 3.0 SSD. Fast, reliable, and survives years of writes.
2. NVMe via HAT ($40-60): For the Pi 5, you can use an NVMe HAT for the fastest possible I/O. Worth it for production setups.

If you must use an SD card initially, upgrade to SSD as soon as you can.

Power Supply: Don’t Cheap Out

This is critical. Many unexplained crashes come from inadequate power. The Pi 5 needs a 5V 5A USB-C supply—the official white power supply from Raspberry Pi. The Pi 4 needs 5V 3A.

Generic phone chargers are NOT sufficient. During AI inference, the Pi can spike to 15-20W, and underpowered supplies cause random reboots. Spend the extra $4 for the official PSU.

Cooling

For 24/7 operation, add active cooling. The Pi 5 throttles at 85°C, and AI workloads generate sustained heat. A cheap case with a 5V fan or the official Raspberry Pi Active Cooler will keep temperatures in the safe zone.

Installation: From Zero to Running

Here’s the streamlined path to get OpenClaw running on your Pi.

Step 1: Flash the OS

Use Raspberry Pi Imager to flash Raspberry Pi OS (64-bit) Lite onto your SD card or SSD. The Lite version is perfect—no desktop environment needed for a headless server.

During setup, configure:

• SSH access (enable it)
• Username and password
• Wi-Fi credentials (or use Ethernet for reliability)

Step 2: Install Node.js

OpenClaw requires Node.js 20 or later. SSH into your Pi and run:

curl -fsSL https://deb.nodesource.com/setup_22.x | sudo bash -
sudo apt install -y nodejs
node --version  # Should show v22.x

Step 3: Run the Installer

OpenClaw provides a one-line installer that handles everything:

curl -sSL https://get.moltbot.org/install-pi.sh | bash

This script:

• Clones the OpenClaw repository
• Installs all dependencies
• Creates a systemd service for auto-start
• Provides pairing instructions

Step 4: Pair Your Device

Start the service and check logs for your pairing code:

sudo systemctl start moltbot
sudo journalctl -u moltbot -f

You’ll see output like:

Your pairing code: ABC-123-DEF
Visit https://get.moltbot.org/pair to complete setup

Visit the URL, enter your code, and your Pi is now connected.

Step 5: Enable on Boot

sudo systemctl enable moltbot

Your AI assistant will now start automatically whenever the Pi boots.

What Works Great on a Pi

Not everything translates well to ARM processors, but OpenClaw’s core features shine on the Pi:

Messaging Channels: Telegram, Discord, Slack, Signal, WhatsApp—all work flawlessly. Your Pi becomes a messaging hub for AI assistance.

Text-Based Skills: Web search, email management, calendar operations, note-taking. These are lightweight and responsive.

Scheduled Tasks: Morning briefings, periodic checks, automated reminders. Set up cron-style triggers and let your Pi handle them 24/7.

Memory System: Persistent conversations, daily notes, long-term context. Your AI remembers across sessions.

Multi-Agent Setups: Run 3-5 specialized agents simultaneously. One for coding, one for research, one for home automation.

What Doesn’t Work (Yet)

Be realistic about the Pi’s limitations:

• Browser Automation: Chromium crawls on ARM. If you need web automation, use the web_fetch tool instead of full browser control.
• Local AI Models (Ollama): The Pi can technically run small models, but inference is painfully slow. Stick to cloud APIs (Claude, GPT, etc.) for now.
• Heavy File Processing: Large PDF analysis, video transcoding—offload these to more powerful machines.
• Apple Ecosystem: iMessage, Apple Notes, and Shortcuts integration requires macOS. If you need these, a Mac mini is your alternative.

Security: A Word of Caution

Running an AI agent with shell access requires careful consideration. Large Language Models can be tricked through prompt injection, potentially leading to unintended command execution.

Best practices:

1. Don’t give OpenClaw access to sensitive APIs or unlimited spending
2. Use dedicated accounts (e.g., a separate GitHub account for the agent)
3. Read and follow the security documentation at docs.openclaw.ai
4. Consider what data the agent can access before connecting it to your personal accounts

This is cutting-edge technology—treat it with appropriate caution.

Performance Optimization Tips

Once you’re running, these tweaks improve stability:

Add Swap Space

sudo dphys-swapfile swapoff
sudo sed -i 's/CONF_SWAPSIZE=.*/CONF_SWAPSIZE=2048/' /etc/dphys-swapfile
sudo dphys-swapfile setup
sudo dphys-swapfile swapon

Log Rotation (prevents disk filling)

sudo tee /etc/logrotate.d/openclaw > /dev/null << 'EOF'
/home/pi/.openclaw/workspace/*.jsonl {
 weekly
 rotate 4
 compress
 missingok
 notifempty
}
EOF

Monitor Temperature

vcgencmd measure_temp

If you’re consistently above 70°C, improve your cooling solution.

The Bottom Line

A Raspberry Pi running OpenClaw isn’t the most powerful AI setup you can build, but it might be the most practical. For roughly $100 in hardware and less than $1 per month in electricity, you get:

• A 24/7 personal AI assistant
• Complete data privacy
• Telegram/Discord/Slack integration
• Automated workflows and scheduled tasks
• Persistent memory and context

It’s an AI assistant for the price of a nice dinner. And unlike a cloud service, you own it completely.

Start with a Pi 5 (8GB), a USB SSD, and the official power supply. Flash the OS, run the installer, pair your device. In under an hour, you’ll have an always-on AI brain living in your home.

Welcome to the future of personal AI infrastructure.