💻 Software Setup Guide

Install Node.js and OpenClaw

OpenClaw is a Node.js application. Node.js 22 LTS provides the best balance of performance and stability on ARM64:

Using npm ci instead of npm install ensures a reproducible installation based on the lockfile. This is the recommended approach for your DIY ClawBox build.

Set Up Ollama — Local LLM Runtime

Ollama is the easiest way to run language models locally. It handles model downloading, quantization, and GPU offloading automatically:

The openclaw AI assistant connects to Ollama's API automatically. You can switch models at any time through the web interface or configuration file.

Configure OpenClaw

Create and customize your configuration file. This is where you define your assistant's personality, capabilities, and integrations:

Key configuration sections for your ClawBox openclaw setup:

• Model: Set defaultModel: "llama3.2:3b" (or your preferred Ollama model). You can define different models for different tasks — a fast model for quick replies, a larger model for complex reasoning.
• Memory: Enable the memory system to let your assistant remember conversations and learn preferences over time. Configure file-based or vector-database-backed persistence.
• Channels: Enable interfaces — web UI on port 3000, Telegram bot, Discord integration, or a local API. Each channel has its own configuration block.
• Plugins: Load skills and plugins for home automation, file management, web browsing, code generation, and more.
• Voice: Configure Piper TTS and Whisper STT for voice interaction. Set preferred voice model and audio devices.

Voice Pipeline Setup

Voice interaction transforms your ClawBox from a text-based tool into a conversational companion. The entire voice pipeline runs locally — no cloud speech services needed:

Configure audio devices in OpenClaw's config file. Most USB audio devices are detected automatically — check with arecord -l for input devices and aplay -l for output.

Set Up Systemd Service

Make your openclaw private AI start automatically on boot and restart if it crashes:

Paste this service definition:

Enable and start:

Performance Tuning

Optimize your system for the best openclaw edge compute performance:

• CPU Governor: Set to "performance" during AI workloads: echo performance | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
• Swap: Disable or minimize swap — swapping kills inference performance. With 16 GB RAM, you likely won't need swap at all.
• GPU Memory: On Pi 5, allocate minimal GPU memory (gpu_mem=16 in /boot/firmware/config.txt) since AI inference runs on CPU.
• Kernel Parameters: Add vm.swappiness=1 and vm.vfs_cache_pressure=50 to /etc/sysctl.d/99-clawbox.conf for better memory management.
• Ollama Tuning: Set OLLAMA_NUM_PARALLEL=1 and OLLAMA_KEEP_ALIVE=5m to avoid memory pressure from concurrent requests.

These optimizations can improve inference speed by 15-30% on the Raspberry Pi 5. For Jetson and x86 systems, additional GPU-specific tuning is available in our Edge Deployment Guide.