The complete guide to hosting your personal AI agent 24/7 on Raspberry Pi 5 (8GB/4GB) and Raspberry Pi 4. Transform your Pi into a dedicated MCP Server with always-on availability, complete privacy, and agentic workflow automation—all for just $1/month in electricity costs.

Why Raspberry Pi?

Running OpenClaw on Raspberry Pi offers incredible advantages for your private AI hosting setup:

24/7 Uptime: Always available AI assistant
Low Power: ~5W average power consumption
Cost Effective: ~$0.43-1.01/month in electricity
100% Private: Your data never leaves your home
Compact Size: Fits anywhere

Value at a Glance

Benefit	Detail
Uptime	24/7 Available
Electricity Cost	~$1/Month
Privacy	100% Private
Initial Cost	~$105 (hardware)

Recommended Hardware Specs for 24/7 Stability

Raspberry Pi Model Comparison

Specification	Raspberry Pi 5 (8GB)	Raspberry Pi 4 (8GB)	Raspberry Pi 4 (4GB)
Performance	Best (2.4x faster)	Excellent	Good
RAM	8GB LPDDR4X-4400	8GB LPDDR4-3200	4GB LPDDR4-3200
Use Case	Heavy local LLM workloads	Multi-user setups	Single-user basic tasks
Recommended	✅ Yes	✅ Yes	⚠️ Minimum viable

Why 8GB RAM is Best for OpenClaw

The 8GB RAM configuration is ideal for running advanced OpenClaw skills and local LLM models because:

Model Context Handling: 8GB allows loading models up to 7B parameters with efficient context processing
Concurrent Skills: Run multiple AI agents simultaneously without memory pressure
Future-Proofing: 2026-era AI models are growing—8GB provides headroom for updates
Swap Performance: More RAM means less swap usage, reducing wear on MicroSD/NVMe storage

For Raspberry Pi 4 4GB users: You can run OpenClaw successfully, but stick to lighter models and single-agent workflows for optimal stability.

⚠️ Critical: Power Supply Requirements for 24/7 AI Servers

Using the correct power supply is essential for preventing OpenClaw crashes during high CPU/NPU load. Many stability issues stem from inadequate power delivery.

Official Raspberry Pi 5 Power Supply Specs (5V 5A USB-C) (Required)

Official Raspberry Pi 5 Power Supply: 5V 5A USB-C (white PSU)

🔌 Don't skimp on the PSU! The Raspberry Pi 5 requires significantly more power than Pi 4, especially during AI inference operations. Third-party power supplies often can't deliver sustained 5A output, causing random reboots when OpenClaw runs heavy workloads.

Why the official 5V 5A matters:

AI workloads spike power draw to 15-20W temporarily
Under-voltage causes immediate service crashes
Official PSU delivers stable power even during agentic workflow execution
Includes a dedicated power cable (not just a generic USB-C cable)

Raspberry Pi 4 Power Supply

Official Raspberry Pi 4 Power Supply: 5V 3A USB-C (black PSU)

While the Pi 4 is more power-efficient, we still recommend the official supply for 24/7 operation. Generic 5V 2.5A supplies may work for desktop use but often fail under continuous AI server loads.

Additional Hardware Requirements

MicroSD card: 32GB+ (A1 or A2 rating recommended)
Case: With active cooling for Pi 5 (recommended for 24/7 operation)
NVMe SSD (Optional but recommended): For best I/O performance and reduced wear

Pro Tip: NVMe SSD for Production Deployments

For serious private AI hosting, consider running Raspberry Pi OS from an NVMe SSD instead of MicroSD. This significantly improves:

I/O Performance: 3-5x faster read/write speeds
Storage Reliability: No SD card corruption risks
Model Loading: Faster local LLM initialization

Running AI models causes frequent disk I/O. While a high-quality microSD card works for testing, an NVMe SSD via the Pi 5's PCIe interface is required for stable 24/7 agentic workflows.

MCP Hub Capabilities: Your 24/7 Home AI Gateway

Your Raspberry Pi running OpenClaw becomes a centralized Model Context Protocol (MCP) gateway for your entire home network. This 2026 architecture pattern enables:

What MCP Hub Enables

Universal AI Access: Any device on your network can query your Pi's AI capabilities via MCP protocol
Skill Orchestration: Chain multiple AI agents across different services (home automation, documents, APIs)
Local Privacy: All inference happens on your Pi—nothing leaves your network
Standardized Interface: MCP compatibility ensures your setup works with future AI tools

Real-World MCP Use Cases

[Smart Home Device] → [OpenClaw MCP Hub on Pi 5] → [Local LLM Inference]
                                                ↓
                                    [Home Automation Execution]

Example workflows:

Natural Language Queries: Ask your smart speaker to query documents stored on your NAS
Automated Workflows: Trigger agentic workflow sequences based on sensor data
Multi-Device Coordination: Your phone, laptop, and smart home all share the same AI brain

Quick Start: One-Line Installer

🚀 The Fastest Path to Your Local AI Server

This single command transforms your Raspberry Pi into a fully-functional MCP Server with agentic workflow capabilities. No manual configuration required—just paste and run.

Security Best Practice

Always inspect scripts before running them. You can view the installer source at github.com/moltbot/install.

One-Line Installer for OpenClaw

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

What happens next:

Clones the OpenClaw repository
Installs all dependencies
Configures the service
Sets up systemd for auto-start
Provides pairing instructions

Step-by-Step Setup Guide

Step 1: Install Raspberry Pi OS (64-bit)

Important: Use 64-bit Lite Version

Download Raspberry Pi OS (64-bit) Lite from the official Raspberry Pi website.

Flashing the OS:

Insert MicroSD card into your computer
Open Raspberry Pi Imager
Choose OS: Raspberry Pi OS (64-bit) Lite
Choose Storage: Select your MicroSD card
Click ⚙️ (Advanced Settings):
- Set hostname: moltbot-pi
- Enable SSH: Use password authentication
- Set username and password
Write and wait for flashing to complete

Headless Setup

By configuring SSH in the imager settings, you can complete the entire setup remotely via SSH—perfect for server deployments without a monitor.

Step 2: Install Dependencies for Your Private AI Hosting

Connect to your Raspberry Pi via SSH and run:

# Update package lists and upgrade system
sudo apt update && sudo apt upgrade -y

# Install Node.js 20.x (using NodeSource repository)
curl -fsSL https://deb.nodesource.com/setup_20.x | sudo -E bash -
sudo apt install -y nodejs

# Verify installations
node --version  # Should be v20.x.x or higher
npm --version   # Should be 10.x.x or higher

# Install Git
sudo apt install -y git

# Install build essentials
sudo apt install -y build-essential python3

Why Node.js 20+? OpenClaw requires Node.js 20 or later for optimal performance and access to the latest JavaScript features essential for modern local LLM operations.

Step 3: Install OpenClaw - Your One-Line Installer for Agentic Workflow

Run the one-line installer to set up your complete MCP Server environment:

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

What the installer does:

Clones the OpenClaw repository
Installs dependencies
Sets up the configuration file
Creates a systemd service for auto-start
Provides pairing instructions

Step 4: Gateway Pairing for Your Private AI Hosting

Complete the pairing process via SSH to activate your local LLM gateway:

# After installation completes, start the OpenClaw service
sudo systemctl start moltbot

# Check service status
sudo systemctl status moltbot

# View logs to see the pairing code
sudo journalctl -u moltbot -f

# You should see output like:
# "Your pairing code: ABC-123-DEF"
# "Visit https://get.moltbot.org/pair to complete setup"

# Keep the service running and visit the pairing URL on your computer
# Enter the code displayed in the logs

# After successful pairing, verify connection
moltbot status

# Enable OpenClaw to start on boot
sudo systemctl enable moltbot

Troubleshooting Pairing Issues

If you see "Gateway Pairing Failed" errors:

Ensure your Raspberry Pi has a stable internet connection
Verify Node.js version is 20+
Check logs with: sudo journalctl -u moltbot -n 50
Visit our troubleshooting guide for more help

Migration Guide: Updating from OpenClaw to OpenClaw

📦 Legacy Migration Box

If you have an existing OpenClaw installation on Linux/Debian, update to the new OpenClaw CLI with these steps:

# Stop the old OpenClaw service
sudo systemctl stop clawdbot

# Navigate to the installation directory
cd ~/clawdbot

# Pull the latest OpenClaw updates
git fetch origin
git checkout main
git pull origin main

# Reinstall dependencies
npm install

# Update systemd service (if service file changed)
sudo cp systemd/moltbot.service /etc/systemd/system/
sudo systemctl daemon-reload

# Start the new OpenClaw service
sudo systemctl start moltbot

# Verify the migration
moltbot --version

# Enable OpenClaw to start on boot
sudo systemctl enable moltbot

What's preserved:

✅ All your existing configurations
✅ Installed skills and extensions
✅ Gateway pairings and credentials
✅ Custom settings and preferences

What's new:

🆕 Enhanced MCP Hub capabilities
🆕 2026-era agentic workflow engine
🆕 Improved local LLM performance
🆕 Expanded hardware compatibility

Performance Optimization

1. Use NVMe SSD (Recommended)

Running Raspberry Pi OS from an NVMe SSD instead of MicroSD significantly improves I/O performance for local LLM operations.

2. Enable Performance Mode

# Edit boot config
sudo nano /boot/firmware/config.txt

# Add these lines for Pi 4:
# arm_freq=1800
# over_voltage=6
# gpu_mem=16

# For Pi 5, the performance mode is enabled by default
# Save and reboot: sudo reboot

3. Configure Swap Space

Increase swap space to 2GB or 4GB for better memory management during AI workloads.

# Edit swapfile size
sudo dphys-swapfile swapoff
sudo nano /etc/dphys-swapfile
# Change CONF_SWAPSIZE=100 to CONF_SWAPSIZE=2048
sudo dphys-swapfile setup
sudo dphys-swapfile swapon

4. Monitor System Resources

# Install htop for monitoring
sudo apt install htop

# Run htop to see CPU, memory, and usage
htop

# Check temperature
vcgencmd measure_temp

# Check clock speeds
vcgencmd measure_clock arm

Why Raspberry Pi 5 Over Mac mini M4?

Choosing the right hardware for your private AI hosting setup depends on your specific needs. Here's why Raspberry Pi 5 excels for dedicated local LLM servers:

Feature	Raspberry Pi 5 (8GB)	Mac mini M4
Initial Cost	~$105 (full setup)	~$599+ (base model)
Monthly Power	~$0.43-1.01/month	~$5-8/month
24/7 Operation	Designed for always-on	General-purpose desktop
Dedicated Device	Runs your AI stack exclusively	Shares resources with macOS
Form Factor	Compact (credit card sized)	Desktop footprint
Upgradability	Expandable storage	Limited
Noise	Silent (passive cooling possible)	Fan noise possible
Use Case	Local LLM server appliance	Daily driver + occasional AI

The Sweet Spot: Raspberry Pi 5 is purpose-built for agentic workflow hosting—low power, dedicated resources, and zero maintenance overhead. Mac mini M4 shines as a primary workstation that can also run AI workloads, but you pay significantly more for that flexibility.

When to Choose Each

Pick Raspberry Pi 5 if: You want a dedicated, always-on MCP Server for pure AI workloads with minimal ongoing costs
Pick Mac mini M4 if: You need a powerful daily workstation that can also run heavy AI models when needed

Cost Analysis

One-Time Hardware Costs

Component	Raspberry Pi 5	Raspberry Pi 4
Board (8GB)	~$80	~$75
Official Power Supply	~$12	~$8
MicroSD card (64GB, A2)	~$12	~$12
Case with cooling	~$15	~$10
Total	~$119	~$105

Monthly Operating Costs

Metric	Value
Power consumption	~5W average
Monthly usage	3.6 kWh/month
@ $0.12/kWh (avg US)	$0.43/month
@ $0.28/kWh (high rate)	$1.01/month

Incredible Value

Compared to running a Mac mini 24/7 (~$5-8/month), a Raspberry Pi costs just $0.43-1/month in electricity. Over a year, that's $5-12 vs $60-96, saving you $48-91 annually. The hardware pays for itself in energy savings within 1-2 years!

Frequently Asked Questions

Does OpenClaw require a cooling fan on Pi 5?

While Raspberry Pi 5 can run passively for light workloads, we strongly recommend active cooling for 24/7 OpenClaw operation. Here's why:

AI Inference Generates Heat: Running local LLM models continuously spikes CPU temperature
Thermal Throttling: Without active cooling, Pi 5 throttles from 2.4GHz to ~1.5GHz under sustained AI loads
Performance Impact: Throttled CPUs process agentic workflow tasks 40% slower
24/7 Reliability: Active cooling ensures consistent performance during long-running operations

Recommended cooling solutions:

Official Raspberry Pi Active Cooler (PID controlled fan)
Case with built-in 5V fan (40mm or larger)
Ensure adequate case ventilation for airflow

Can I run OpenClaw on Raspberry Pi 4 4GB?

Yes, OpenClaw can run on Raspberry Pi 4 with 4GB RAM, making it an excellent entry-level option for private AI hosting. However, for optimal performance with heavier local LLM workloads and complex agentic workflow tasks, we recommend 8GB RAM. The 4GB model works well for:

Basic MCP Server operations
Text-based AI interactions
Single-user scenarios
Lightweight skill execution

Tip: If you're planning to run multiple concurrent AI agents or process larger contexts, the 8GB model provides significantly better headroom and future-proofs your setup for 2026-era AI models.

What is the best power supply for a 24/7 AI server?

The best power supply for running OpenClaw 24/7 on Raspberry Pi is the official Raspberry Pi foundation power supply:

Raspberry Pi 5: Official 5V 5A USB-C power supply (white)
Raspberry Pi 4: Official 5V 3A USB-C power supply (black)

⚠️ Critical Warning: Avoid third-party or generic USB-C power supplies for 24/7 AI server duty. Many can't sustain the amperage needed during AI inference spikes, causing random reboots and service crashes that are difficult to diagnose.

Why official supplies matter for AI workloads:

AI model inference causes brief power draw spikes (15-20W on Pi 5)
Under-voltage triggers immediate CPU throttling or system shutdown
Official PSUs deliver stable, consistent power even under variable loads
Dedicated power cables (not generic USB-C) ensure proper current delivery

Cost vs. Reliability: At ~$12 for the official Pi 5 PSU vs ~$8 for generic, the $4 difference prevents hours of troubleshooting stability issues.

Is Pi 5 faster than M4 for simple automations?

For simple automations and basic agentic workflow tasks, Raspberry Pi 5 offers surprising advantages over Mac mini M4:

Where Pi 5 Wins:

Dedicated Resources: Pi 5 runs only your AI stack, while M4 shares resources with macOS and background processes
Zero Contention: No competing applications stealing CPU cycles during automation execution
Instant Availability: Always-on, no wake-from-sleep delays
Network Isolation: Dedicated device means predictable response times

Benchmark Comparison (Simple Text Automation):

Raspberry Pi 5: ~200-400ms response time (dedicated)
Mac mini M4: ~150-300ms response time (when awake, not under load)

The Reality: While M4 has faster raw specs, real-world simple automation performance is comparable because:

Network latency often dominates the response time
Pi 5's dedicated architecture eliminates resource contention
Most automations are I/O bound, not CPU bound

Recommendation: Choose Pi 5 for dedicated 24/7 automation duty, M4 if you need a workstation that can also run automations during active use.

Next Steps

Once your OpenClaw setup is complete, expand your Pi's capabilities:

Browse 338+ AI Skills - Add powerful new capabilities to your MCP Server
Basic Configuration - Customize your OpenClaw settings
Troubleshooting Guide - Get help with common issues

Need Help?

Having trouble with your Raspberry Pi setup?

Quick Reference

Essential Commands

# Start OpenClaw
sudo systemctl start moltbot

# Check status
sudo systemctl status moltbot

# View logs
sudo journalctl -u moltbot -f

# Restart service
sudo systemctl restart moltbot

# Enable auto-start on boot
sudo systemctl enable moltbot

# Check version
moltbot --version

Performance Monitoring

# Check CPU temperature
vcgencmd measure_temp

# Check clock speed
vcgencmd measure_clock arm

# Monitor resources
htop

# Check disk space
df -h

Power Supply Quick Check

# Check for under-voltage events (indicates weak PSU)
vcgencmd get_throttled

# Output meanings:
# 0x0 = No throttling (good PSU)
# 0x50005 = Under-voltage occurred (replace PSU immediately)

Technical Specifications

Minimum Requirements

CPU: ARM Cortex-A72 (4 cores)
RAM: 4GB (8GB+ recommended)
Storage: 32GB MicroSD (NVMe SSD recommended)
Network: Ethernet or Wi-Fi
Power: Official 5V 5A (Pi 5) or 5V 3A (Pi 4) USB-C supply

Recommended Specifications

CPU: ARM Cortex-A76 (4 cores) - Raspberry Pi 5
RAM: 8GB
Storage: 256GB NVMe SSD
Network: Ethernet (preferred)
Power: Official Raspberry Pi 5V 5A USB-C power supply
Cooling: Active cooler for 24/7 operation

Last Updated: January 2026 | OpenClaw v2.0 (formerly Clawdbot) Compatible: Raspberry Pi 4 (4GB/8GB), Raspberry Pi 5 (8GB) OS: Raspberry Pi OS (64-bit) Lite or Desktop Perfect for: Local LLM deployment, MCP Server hosting, Agentic Workflow automation, 24/7 AI server, Private AI hosting

OpenClaw Raspberry Pi 5 & 4 Setup Guide | 2026 Private AI Server