Build a $50 always-on agent on a Raspberry Pi.

A Raspberry Pi is the cheapest realistic OpenClaw host that isn't a free-tier VPS. It's also the most fun one to build — there's something delightful about a blinking little box on your shelf running a fully-functional AI agent.

The Pi is a great fit for: a personal assistant on Telegram or WhatsApp, a daily digest agent, a cron-driven monitoring agent. It's a poor fit for: heavy browser automation, multi-agent setups, anything that runs a local model larger than 3B.

Total damage: $90–$130 depending on extras you want. None of these links are affiliated; pick whichever vendor has stock.

The official Raspberry Pi Imager is the cleanest path. Install it on your laptop, plug in the SD card, and configure everything before the Pi boots for the first time.

• Choose 'Raspberry Pi OS Lite (64-bit)' — no desktop needed
• Click the gear icon: enable SSH (key-based), set hostname 'openclaw', set username 'openclaw', paste your SSH public key
• Set Wi-Fi if you must, but prefer Ethernet
• Flash, insert SD card into Pi, plug in Ethernet, plug in power
• Find the Pi on your network: `ssh openclaw.local` should work via mDNS

After first boot, switch to the SSD as the boot drive — SD cards aren't built for OpenClaw's write patterns:

# Once booted, copy the SD to the SSD and switch
sudo rpi-clone sda
# Then in raspi-config: Advanced → Boot Order → USB first
sudo raspi-config
sudo reboot

OpenClaw's official packages include ARM64 binaries. No special flags needed.

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

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

# Verify ARM detection
node --version  # v22.x.x
uname -m        # aarch64

# Install OpenClaw
curl -fsSL https://openclaw.ai/install-cli.sh | bash
openclaw onboard

Most Pi setups stick with cloud LLMs and use the Pi just as the always-on host. If you do want local inference, your options narrow to the smallest models.

Reality check: even the best Pi-runnable model is materially worse than Claude Haiku on the API. The math usually favors "use Haiku via API at $0.10 per 1M tokens" over "run a 3B locally on the Pi." Local-on-Pi only wins if you're optimizing for offline operation or strict data locality.

Headless Chrome on a Pi is technically supported and practically painful. Page rendering is 5–20x slower than on a real CPU, and a single Chrome process eats 800 MB+ of RAM — which on a Pi 4 4 GB leaves nothing for the gateway and the agent.

The right answer if you need browser automation: run the browser elsewhere.

{
  "browser": {
    "mode": "remote",
    "cdpUrl": "ws://browser-vps.tail-net.ts.net:9222"
  }
}

Spin up a 2 GB VPS just for headless Chrome (browserless/chrome container), put it on Tailscale, point your Pi-hosted agent at it. Best of both worlds: cheap always-on home server + real browser performance for the rare browsing.

Pis are stable, but home environments aren't. Three uptime wins:

• Run as a systemd service. openclaw service install — same as on a VPS.
• Add a UPS. A $40 mini UPS lets the Pi survive 5–15 minutes of power outage. Critical if your agent handles anything time-sensitive.
• Watchdog timer. Enable the hardware watchdog in /boot/firmware/config.txt with dtparam=watchdog=on — the Pi reboots itself if it locks up.
• Healthcheck pings. Same Healthchecks.io pattern as a VPS: ping every 5 min, get an email if it goes silent.