What You Need to Know
Raspberry Pi 4 (4GB+) or Pi 5 is the minimum recommended hardware. The gateway and Node.js runtime need about 500MB-1GB of RAM during normal operation, leaving limited headroom for local inference. For this reason, external API providers (Anthropic, OpenAI, Google) are strongly recommended over local models like Ollama on Pi hardware.
Install Node.js 22+ using the official ARM builds or nvm. The standard OpenClaw installer works on ARM Linux: curl -fsSL https://openclaw.ai/install.sh | bash. Run onboarding with --install-daemon to register a systemd service, which ensures the gateway starts on boot and restarts after crashes.
Choose a lightweight provider configuration. Set your primary model to a cost-efficient option and use heartbeat.model override to run heartbeat checks on the cheapest available model. Keep automation frequency conservative — running heartbeat every 30 minutes with a cheap model is more sustainable than every 5 minutes with a premium model on constrained hardware.
Monitor storage, swap, and memory usage from day one. Pi SD cards have limited write endurance, so configure log rotation and session cleanup (openclaw sessions cleanup) aggressively. If you are running memory indexing, the SQLite vector database can grow significantly over months — monitor disk usage and consider mounting an external SSD for the ~/.openclaw directory.
For remote reliability, harden your network access with SSH keys, firewall rules, and ideally a VPN like Tailscale or WireGuard. Set up health monitoring that alerts you if the Pi goes offline. Validate health endpoints, log retention, and backup strategy before trusting long-running automations. For many users, a $5/month VPS gives better price/performance than a Pi for 24/7 assistant use — but the Pi wins for offline-capable, privacy-first, or smart-home-adjacent deployments.