Meta Description: Install OpenClaw on Raspberry Pi 4/5 for a 24/7 AI agent. Thermal management, swap config, headless boot, ARM compatibility, and when to upgrade to cloud hosting explained.
A $35 computer running your AI agent 24/7. No cloud bills. No VPS subscription. Just a Raspberry Pi on your desk, always listening, always ready.
The Pi 4 and Pi 5 are powerful enough to run OpenClaw for personal use. 4GB of RAM handles most skills. Low power consumption means it costs pennies per month to run. And it’s yours—no one else’s server, no data leaving your network.
This guide covers Pi 4/5 installation with thermal considerations, swap configuration for memory-constrained environments, headless boot setup, and which skills are too heavy for ARM architecture.
We’ll also be honest about when a Pi isn’t enough and you should upgrade to cloud hosting instead.
Why Raspberry Pi for OpenClaw
The case for Pi:
Cost: $35-75 depending on model and RAM. One-time purchase vs $5-20/month VPS.
Privacy: Your data stays on your network. No cloud provider can access it.
Always-on: Runs 24/7 for ~$2/month in electricity. Compare to laptop ($$) or desktop ($$$).
Learning: Physical server you can touch. See the activity LEDs. Power cycle by unplugging.
Edge computing: Responds instantly (no internet round-trip). Works during outages if you have local power.
The tradeoffs:
Limited RAM: 4GB or 8GB max (Pi 5). Some skills need more.
ARM architecture: Not all Node packages have ARM builds. Expect compatibility issues.
SD card reliability: MicroSD fails eventually. Requires backup discipline.
Thermal throttling: Gets hot under load. Needs cooling solution.
No redundancy: If Pi dies, agent is down. No automatic failover.
When Pi makes sense:
- Personal AI assistant (1 user)
- Low-moderate message volume
- Primarily text-based interactions
- Home network deployment
- Learning/experimentation
When Pi doesn’t make sense:
- Team/multi-user deployment
- High message volume (>100/day)
- Heavy skills (image generation, voice processing)
- Mission-critical uptime needs
- Public internet exposure
Hardware Requirements
Recommended Setup
Raspberry Pi 5 (8GB) – Current generation, best performance
- $80 for 8GB model
- ARM Cortex-A76 (quad-core, 2.4GHz)
- 2× USB 3.0, 2× USB 2.0
- Gigabit Ethernet + WiFi 6
- PCIe for NVMe (optional)
Raspberry Pi 4 (4GB or 8GB) – Previous gen, still solid
- $35-55 depending on RAM
- ARM Cortex-A72 (quad-core, 1.5GHz)
- 2× USB 3.0, 2× USB 2.0
- Gigabit Ethernet + WiFi 5
Minimum: Pi 4 with 4GB RAM. Anything less will struggle.
Required Accessories
MicroSD card: 32GB minimum, 64GB recommended
- SanDisk Extreme or Samsung EVO
- NOT generic brands (high failure rate)
Power supply: Official Pi power supply
- 5V/3A for Pi 4
- 5V/5A for Pi 5 (uses USB-C PD)
- Don’t use phone chargers (causes stability issues)
Cooling solution: Essential for sustained load
- Minimum: Heatsinks (passive cooling)
- Better: Case with fan (active cooling)
- Best: Argon ONE case (metal case acts as giant heatsink + fan)
Optional but recommended:
- Ethernet cable (more stable than WiFi)
- Case (protect from dust/damage)
- Backup MicroSD or USB drive
Initial Setup
Flash Raspberry Pi OS
Download Raspberry Pi Imager: raspberrypi.com/software
Choose OS:
- Raspberry Pi OS Lite (64-bit) – Command-line only, minimal resources
- NOT Desktop version (wastes RAM on GUI)
Configure before flash:
Click gear icon for advanced options:
- Set hostname: openclaw
- Enable SSH: Yes
- Set username/password: pi / [your password]
- Configure WiFi: SSID + password (if using WiFi)
- Set locale: Your timezone
Flash to MicroSD:
- Insert card into computer
- Select OS and card
- Write
First boot:
- Insert MicroSD into Pi
- Connect ethernet (or rely on WiFi config)
- Power on
- Wait 60 seconds for boot
Find Pi’s IP:
From your computer:
ping openclaw.local
Or check your router’s DHCP list.
SSH into Pi:
Enter password you configured.
You’re in.
System Preparation
Update System
sudo apt update
sudo apt upgrade -y
Install Node.js (ARM-compatible)
Don’t use apt’s Node.js (old version).
Use NodeSource repository:
curl -fsSL https://deb.nodesource.com/setup_20.x | sudo -E bash -
sudo apt install -y nodejs
Verify:
node --version # Should show v20.x
npm --version # Should show 10.x
Install Git
sudo apt install -y git
Install SQLite (for memory backend)
sudo apt install -y sqlite3
Configure Swap (Critical for 4GB Models)
Pi 4 with 4GB RAM will run out of memory under load. Swap extends RAM using SD card storage.
Check Current Swap
free -h
Default: 100MB swap (not enough).
Increase Swap to 2GB
Edit swap config:
sudo nano /etc/dphys-swapfile
Find line:
CONF_SWAPSIZE=100
Change to:
CONF_SWAPSIZE=2048
Save: Ctrl+X, Y, Enter
Restart swap:
sudo dphys-swapfile swapoff
sudo dphys-swapfile setup
sudo dphys-swapfile swapon
Verify:
free -h
Should show 2GB swap.
Important: Swap on MicroSD reduces card lifespan. Use high-quality card and expect to replace every 2-3 years.
Install OpenClaw
Clone Repository
cd ~
git clone https://github.com/openclaw/openclaw.git
cd openclaw
Install Dependencies
npm install
This takes 10-15 minutes on Pi. ARM compilation is slow.
If you see errors about native modules:
Some packages don’t have ARM builds. Common issues:
better-sqlite3: Usually works but needs compilation
npm rebuild better-sqlite3
sharp (image processing): Works on ARM but slow to compile
npm rebuild sharp
If package fails completely: You’ll need to disable that skill.
Create Configuration
cp .env.example .env
nano .env
Configure:
# Core
NODE_ENV=production
# Telegram
TELEGRAM_ENABLED=true
TELEGRAM_BOT_TOKEN=your-bot-token
TELEGRAM_PHONE_NUMBER=+15551234567
# WhatsApp (works but drains resources)
WHATSAPP_ENABLED=false # Enable only if needed
# Memory
MEMORY_BACKEND=sqlite
MEMORY_DB_PATH=./data/memory.db
# Logging
LOG_LEVEL=info
WhatsApp note: WhatsApp uses Chrome/Chromium in background. Heavy on Pi. Enable only if you need it.
Create Data Directory
mkdir -p data logs
Test Run
npm start
Should see:
[INFO] OpenClaw starting...
[INFO] Telegram bot connected
[INFO] Agent ready
Ctrl+C to stop.
Run as System Service
Make OpenClaw auto-start on boot and restart on crash.
Create Systemd Service
sudo nano /etc/systemd/system/openclaw.service
Paste:
[Unit]
Description=OpenClaw AI Agent
After=network.target
[Service]
Type=simple
User=pi
WorkingDirectory=/home/pi/openclaw
ExecStart=/usr/bin/node index.js
Restart=always
RestartSec=10
StandardOutput=append:/home/pi/openclaw/logs/service.log
StandardError=append:/home/pi/openclaw/logs/service-error.log
# Environment
Environment=NODE_ENV=production
[Install]
WantedBy=multi-user.target
Save: Ctrl+X, Y, Enter
Enable and Start Service
sudo systemctl daemon-reload
sudo systemctl enable openclaw
sudo systemctl start openclaw
Check status:
sudo systemctl status openclaw
Should show “active (running)”.
View logs:
tail -f ~/openclaw/logs/service.log
Service Management Commands
sudo systemctl start openclaw # Start
sudo systemctl stop openclaw # Stop
sudo systemctl restart openclaw # Restart
sudo systemctl status openclaw # Check status
Thermal Management
Pi throttles CPU when temperature hits 80°C. This makes OpenClaw sluggish.
Check Temperature
vcgencmd measure_temp
Shows current temp (e.g., temp=65.0'C).
Safe range: 40-75°C Warning: 75-80°C Throttling: 80°C+
Monitor While Running
watch -n 2 vcgencmd measure_temp
Updates temp every 2 seconds.
Cooling Solutions
Level 1: Heatsinks (Passive)
$5 aluminum heatsinks on CPU and RAM chips.
- Lowers temp by 5-10°C
- No noise
- Insufficient under sustained load
Level 2: Case with Fan (Active)
$10-15 case with 30mm fan.
- Lowers temp by 15-20°C
- Slight noise
- Sufficient for most use cases
Level 3: Argon ONE Case
$25 metal case that IS the heatsink + controllable fan.
- Lowers temp by 20-30°C
- Quietest fan (only spins when needed)
- Best solution for 24/7 operation
Pi 5 specific: Runs hotter than Pi 4. Active cooling mandatory.
If You’re Throttling
Reduce CPU usage:
Edit service to use less Node threads:
Environment=UV_THREADPOOL_SIZE=2
Limit agent skills:
Disable heavy skills in config:
SKILLS_DISABLED=image-gen,video-processing,voice-synthesis
Add fan immediately.
Skills That Are Too Heavy for ARM/Pi
Not all OpenClaw skills run well on Pi. Some are too CPU/RAM intensive, others lack ARM support.
Skills That Work Fine
✅ Text processing – Summarization, Q&A, general chat ✅ Telegram/Discord – Messaging integration ✅ Calendar integration – Google Calendar API ✅ File operations – Read/write documents ✅ Web scraping – Fetching data from websites ✅ SQLite operations – Database queries
Skills That Are Slow But Functional
⚠️ Image analysis – Works but takes 30-60 seconds per image ⚠️ PDF processing – Large PDFs (>100 pages) are slow ⚠️ WhatsApp – Runs Chrome in background, uses 500MB+ RAM
Skills That Don’t Work or Cause Problems
❌ Voice synthesis (TTS) – Most TTS engines lack ARM builds ❌ Speech recognition (STT) – Too CPU intensive, real-time impossible ❌ Image generation – Stable Diffusion etc. need GPU ❌ Video processing – Encoding/decoding too slow ❌ Large language model inference – Need 8GB+ VRAM (Pi has 0)
If you need these skills: Cloud deployment required (VPS or PaioClaw).
Testing Skill Performance
# Enable debug logging
export LOG_LEVEL=debug
# Test specific skill
npm run test-skill image-analysis
If execution takes >5 seconds or causes temperature spike, skill is too heavy.
Headless Operation (No Monitor)
Pi runs without keyboard/monitor/display. Access via SSH or web UI.
SSH Access from Anywhere
Problem: Home IP changes, can’t SSH from outside network.
Solution 1: Dynamic DNS
Use service like NoIP or DuckDNS:
- Register free hostname (e.g., myclaw.ddns.net)
- Install update client on Pi
- Hostname always points to your home IP
Solution 2: Tailscale (Recommended)
Zero-config VPN. Access Pi from anywhere.
curl -fsSL https://tailscale.com/install.sh | sh
sudo tailscale up
Now SSH via Tailscale hostname (stays same even when home IP changes).
Web UI (Optional)
If OpenClaw has web interface:
Forward port in router: 3000 → Pi’s IP
Access: http://your-home-ip:3000
Security warning: Don’t expose to internet without auth.
Monitoring
Install monitoring tool:
sudo apt install -y htop
View system resources:
htop
Shows CPU, RAM, running processes.
Backup Strategy
MicroSD cards fail. Backup regularly or lose everything.
Daily Automated Backup
Backup script:
Create ~/backup.sh:
#!/bin/bash
BACKUP_DIR=/home/pi/backups
DATE=$(date +%Y%m%d)
mkdir -p $BACKUP_DIR
# Backup OpenClaw data
tar -czf $BACKUP_DIR/openclaw-$DATE.tar.gz
/home/pi/openclaw/data
/home/pi/openclaw/.env
/home/pi/openclaw/config
# Keep only last 7 days
find $BACKUP_DIR -name "openclaw-*.tar.gz" -mtime +7 -delete
Make executable:
chmod +x ~/backup.sh
Schedule daily at 2am:
crontab -e
Add:
0 2 * * * /home/pi/backup.sh
Weekly Manual Backup
Stop agent:
sudo systemctl stop openclaw
Backup to external USB:
sudo rsync -av /home/pi/openclaw /media/usb-drive/openclaw-backup
Restart agent:
sudo systemctl start openclaw
SD Card Cloning (Full System Backup)
Use Pi Imager to create image of entire SD card.
- Shut down Pi
- Remove SD card
- Insert into computer
- Raspberry Pi Imager → “Read” mode
- Save .img file
If card fails, flash .img to new card = instant recovery.
Power Management
Pi has no power button. Power cycle = unplug/replug.
Graceful Shutdown
From SSH:
sudo shutdown now
Wait 30 seconds for Pi to power off (LED stops blinking).
Then unplug power.
Never just yank power while Pi is running. Corrupts SD card.
UPS for Pi
$30 battery backup (Geekworm/Waveshare UPS HAT).
Provides:
- 30-60 min backup power
- Clean shutdown on power loss
- Prevents data corruption
Worth it for 24/7 operation.
Power Consumption
Pi 4 idle: 3-4W Pi 4 under load: 6-8W
Cost at $0.15/kWh:
- Idle: $0.40/month
- 24/7 under load: $0.80/month
Compare to:
- Laptop 24/7: $10-15/month
- Desktop 24/7: $30-50/month
- Cloud VPS: $5-20/month
When to Upgrade from Pi to Cloud
Pi is amazing for learning and personal use. But there are clear limits.
You’ve Outgrown Pi When:
Performance:
- Agent response time >5 seconds regularly
- CPU at 100% constantly
- Temperature throttling frequently
- Out of memory errors
Reliability:
- SD card failed (2nd or 3rd time)
- Need redundancy/failover
- Can’t afford downtime
Scale:
- Multiple users need access
- Message volume >100/day
- Need skills Pi can’t run (voice, images, video)
Connectivity:
- Need guaranteed uptime
- Home internet too unreliable
- Want access from anywhere without VPN
Cloud Alternatives
Budget VPS ($5-10/month):
- DigitalOcean, Linode, Vultr
- 2GB+ RAM, real CPU
- 99.9% uptime SLA
- Professional networking
When to choose: Personal use, moderate scale, want control.
PaioClaw Managed ($15-25/month):
- No server management
- Automatic scaling
- Built-in monitoring
- Professional support
When to choose: Don’t want to manage servers, need reliability, team use.
The Honest Take
Pi is perfect for:
- Learning OpenClaw
- Personal AI assistant (1 user)
- Home automation integration
- Privacy-first deployment
- Experimentation
Pi struggles with:
- Multiple users
- Heavy skills (voice, images, video)
- Mission-critical uptime
- High message volume
- Professional/business use
Pi costs:
- Hardware: $35-150 (one-time)
- Electricity: $0.40-0.80/month
- SD card replacement: $10/year
Cloud costs:
- VPS: $5-20/month ongoing
- PaioClaw: $15-25/month ongoing
The decision: If Pi handles your workload and you enjoy tinkering, it’s unbeatable value. If you’re fighting thermal issues, memory limits, or ARM incompatibility, cloud is cheaper than your time.
The Bottom Line
Raspberry Pi can absolutely run OpenClaw for personal use. The Pi 5 with 8GB RAM handles text-based agent tasks smoothly. Add proper cooling and a good SD card, you’ve got a reliable $35 AI agent running 24/7.
The limitations are real: some skills won’t work, heavy workloads cause throttling, and SD cards eventually fail. But for learning, experimentation, or personal use, Pi delivers exceptional value.
If you followed this guide, you now have OpenClaw running on a Pi with proper thermal management, swap configuration, and automatic startup.
Whether you stick with Pi or upgrade to cloud depends on your needs. Pi for personal/learning. Cloud for production/team use.
Ready to skip the hardware entirely? PaioClaw runs OpenClaw in the cloud with zero setup, automatic scaling, and professional monitoring. No Pi to configure, no SD cards to fail, no thermal management. Starts FREE, Smart $15/month, Genius $25/month. Start free →