Running OpenClaw on Raspberry Pi: Complete Setup Guide

A Raspberry Pi 4 can run a fully functional AI agent 24/7 for under $3/month in electricity. Here's exactly how to set it up.

---

Table of Contents

1. Prerequisites
2. What You'll Get
3. Hardware & Software Requirements
4. Step-by-Step Setup
5. Cost Optimization
6. Troubleshooting
7. FAQ

---

Prerequisites

Before starting, make sure you have:

• Raspberry Pi 4 (4GB or 8GB RAM)
• MicroSD card (32GB+) or USB SSD
• Power supply (official 5V/3A recommended)
• Network connection (Ethernet preferred)
• API key from OpenAI or Anthropic
• Basic terminal/SSH knowledge

Time required: ~30 minutes

---

What You'll Get

flowchart LR
    subgraph Pi["Raspberry Pi"]
        OC[OpenClaw Gateway]
        LLM[GPT-5-nano / Ollama]
    end

    subgraph Channels["Communication"]
        DC[Discord]
        TG[Telegram]
        WA[WhatsApp]
    end

    subgraph Features["Capabilities"]
        F1[24/7 Availability]
        F2[Scheduled Tasks]
        F3[Proactive Alerts]
        F4[Offline Fallback]
    end

    Pi <--> Channels
    Pi --> Features

By the end of this guide:

• OpenClaw running on your Pi
• Connected to Discord/Telegram/WhatsApp
• Using cheap models by default (pennies per query)
• Local LLM fallback via Ollama
• Scheduled jobs and monitoring
• Production-ready configuration

---

Requirements

Hardware

| Component | Minimum | Recommended | |-----------|---------|-------------| | Pi Model | Pi 4 (4GB) | Pi 4 (8GB) or Pi 5 | | Storage | 32GB SD | 64GB+ SD or SSD | | Power | Official 5V/3A | Official PSU | | Network | WiFi | Ethernet (more reliable) |

Software

• Raspberry Pi OS (64-bit) — Bookworm or newer
• Node.js 20+
• Git

Accounts

• API key for at least one provider (OpenAI, Anthropic, etc.)
• Optional: Discord bot token, Telegram bot token

---

Step 1: Prepare the Pi

Flash the OS

1. Download Raspberry Pi Imager
2. Choose: Raspberry Pi OS (64-bit)
3. Configure: Enable SSH, set username/password, WiFi
4. Flash to SD card

First Boot

# SSH into your Pi
ssh [email protected]

# Update everything
sudo apt update && sudo apt upgrade -y

# Install essentials
sudo apt install -y git curl build-essential

Install Node.js

# Install Node.js 22 via NodeSource
curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
sudo apt install -y nodejs

# Verify
node --version  # Should show v22.x.x
npm --version

---

Step 2: Install OpenClaw

# Install globally
sudo npm install -g openclaw

# Verify installation
openclaw --version

Run Setup Wizard

openclaw setup

The wizard will ask:

1. Mode: Choose local (runs on the Pi itself)
2. Primary model: Recommend gpt-5-nano for cost efficiency
3. API key: Enter your OpenAI or Anthropic key
4. Channels: Configure Discord/Telegram if desired

---

Step 3: Configure for Production

Optimize for Pi Resources

Edit ~/.openclaw/openclaw.json:

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "openai/gpt-5-nano",
        "fallbacks": [
          "openai/gpt-5-mini",
          "ollama/mistral:7b"
        ]
      },
      "contextPruning": {
        "mode": "cache-ttl",
        "ttl": "3m"
      },
      "maxConcurrent": 2
    }
  }
}

Why These Settings?

| Setting | Value | Reason | |---------|-------|--------| | gpt-5-nano | Primary | $0.05/1M tokens — dirt cheap | | maxConcurrent: 2 | Low | Pi has limited RAM | | ttl: 3m | Short | Aggressive memory cleanup | | Ollama fallback | Local | Works offline, free |

---

Step 4: Add Local LLM (Optional but Recommended)

flowchart TD
    REQ[Request] --> CHECK{API Available?}
    CHECK -->|Yes| API[Use GPT-5-nano]
    CHECK -->|No| OLLAMA[Use Ollama Local]
    API --> RESP[Response]
    OLLAMA --> RESP

Ollama lets you run models locally — no API costs, works offline.

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull a small, fast model
ollama pull mistral:7b  # ~4GB, runs well on 8GB Pi

# Or even smaller
ollama pull phi3:mini   # ~2GB, faster but less capable

Now if your API fails or you want free queries, Ollama handles them.

---

Step 5: Set Up Channels

Discord

1. Create a bot at Discord Developer Portal
2. Get your bot token
3. Add to config:

{
  "channels": {
    "discord": {
      "enabled": true,
      "token": "YOUR_BOT_TOKEN",
      "dm": {
        "policy": "allowlist",
        "allowFrom": ["YOUR_USER_ID"]
      }
    }
  }
}

Telegram

1. Create a bot via @BotFather
2. Get your bot token
3. Add to config:

{
  "channels": {
    "telegram": {
      "botToken": "YOUR_BOT_TOKEN",
      "dmPolicy": "allowlist",
      "allowFrom": ["YOUR_TELEGRAM_ID"]
    }
  }
}

---

Step 6: Run as a Service

You want OpenClaw to start on boot and restart on crash.

Create systemd Service

sudo nano /etc/systemd/system/openclaw.service

[Unit]
Description=OpenClaw AI Agent
After=network.target

[Service]
Type=simple
User=pi
WorkingDirectory=/home/pi
ExecStart=/usr/bin/openclaw gateway start --foreground
Restart=always
RestartSec=10
Environment=NODE_ENV=production

[Install]
WantedBy=multi-user.target

Enable and Start

sudo systemctl daemon-reload
sudo systemctl enable openclaw
sudo systemctl start openclaw

# Check status
sudo systemctl status openclaw

# View logs
journalctl -u openclaw -f

---

Step 7: Cost Optimization

Model Pricing (Feb 2026)

pie title Cost per 1M Tokens (Input)
    "gpt-5-nano ($0.05)" : 5
    "gpt-5-mini ($0.15)" : 15
    "claude-haiku ($0.25)" : 25
    "Ollama (FREE)" : 0

| Model | Input | Output | Best For | |-------|-------|--------|----------| | gpt-5-nano | $0.05/1M | $0.15/1M | Default, cheap | | gpt-5-mini | $0.15/1M | $0.60/1M | Better quality | | claude-haiku-4 | $0.25/1M | $1.25/1M | Anthropic cheap | | ollama/mistral | Free | Free | Offline/backup |

Estimated Monthly Costs

| Usage Level | Queries/Day | Monthly Cost | |-------------|-------------|--------------| | Light | 20 | ~$0.50 | | Moderate | 100 | ~$2.50 | | Heavy | 500 | ~$12.00 |

Plus electricity: ~$2-3/month for 24/7 Pi operation.

---

Step 8: Add Scheduled Jobs

flowchart LR
    CRON[Cron Scheduler] --> J1[Morning Briefing 9AM]
    CRON --> J2[Portfolio Check 6PM]
    CRON --> J3[Weekly Summary Sun]

    J1 --> AGENT[OpenClaw Agent]
    J2 --> AGENT
    J3 --> AGENT

    AGENT --> NOTIFY[Send to Discord/Telegram]

OpenClaw can run tasks on a schedule:

{
  "cron": {
    "enabled": true,
    "jobs": [
      {
        "name": "morning-briefing",
        "schedule": { "kind": "cron", "expr": "0 9 * * *" },
        "payload": {
          "kind": "agentTurn",
          "message": "Give me a morning briefing"
        }
      }
    ]
  }
}

---

Step 9: Monitoring & Maintenance

Health Check Script

#!/bin/bash
# /home/pi/check-openclaw.sh

if ! systemctl is-active --quiet openclaw; then
    echo "OpenClaw is down, restarting..."
    sudo systemctl restart openclaw
fi

Add to crontab:

*/5 * * * * /home/pi/check-openclaw.sh

---

Troubleshooting

"Out of Memory" Errors

# Add swap space
sudo fallocate -l 2G /swapfile
sudo chmod 600 /swapfile
sudo mkswap /swapfile
sudo swapon /swapfile

# Make permanent
echo '/swapfile none swap sw 0 0' | sudo tee -a /etc/fstab

Ollama Too Slow

Use a smaller model:

ollama pull phi3:mini      # Faster, less capable
ollama pull tinyllama      # Even smaller

API Timeouts

Increase timeout in config:

{
  "gateway": {
    "timeout": 120000
  }
}

Can't Connect to Discord/Telegram

1. Check your bot token is correct
2. Verify the Pi has internet access
3. Check firewall isn't blocking outbound

---

Performance Tips

| Tip | Impact | |-----|--------| | Use Ethernet over WiFi | More reliable, lower latency | | Use SSD over SD card | Faster, longer lifespan | | Disable Bluetooth | Free up resources | | Add swap space | Prevent OOM crashes |

# Disable unnecessary services
sudo systemctl disable bluetooth
sudo systemctl disable avahi-daemon

---

What Can Your Pi Agent Do?

Yes

• Answer questions via Discord/Telegram 24/7
• Run scheduled tasks (reminders, briefings)
• Monitor things and alert you
• Perform web searches
• Execute code (carefully!)
• Work offline with local LLMs

Not Well

• Heavy computation (use a beefier machine)
• Image generation (no GPU)
• Real-time voice (too slow)

---

FAQ

How much does it cost to run?

~$3-5/month total: $2-3 electricity + $0.50-2.50 API costs depending on usage.

Can I use Claude instead of GPT?

Yes! Use anthropic/claude-haiku-4 as your primary model. It's slightly more expensive but some prefer it.

Will it work on Pi 3?

Technically yes, but not recommended. Pi 3 has only 1GB RAM and will struggle with even small Ollama models.

Can I run multiple agents?

Yes! Check our multi-agent guide for running agents across Pi and Mac.

What if my internet goes down?

With Ollama configured as fallback, your agent continues working offline with local models.

How do I update OpenClaw?

sudo npm update -g openclaw
sudo systemctl restart openclaw

---

Next Steps

1. Add more channels — WhatsApp, Slack, email
2. Install skills — Weather, GitHub, crypto prices
3. Set up multi-agent — Connect Pi to main machine
4. Add tools — Web scraping, API integrations

---

Resources

• OpenClaw Documentation
• OpenClaw GitHub
• OpenClaw Discord
• ClawHub Skills

---

Related Posts

• How to Build a Multi-Agent AI System: Mac + Raspberry Pi
• How to Run Multiple AI Bots in Discord
• AI Agents vs ChatGPT Wrappers

---

About the Author

BlestLabs builds AI-powered tools and agents. We run a Pi-based agent (Pip) 24/7 alongside our Mac Mini agent (Alfred) for research and automation. Follow our journey on Twitter @aceism_.