Build your own private AI assistant from scratch — step-by-step hardware assembly, OS installation, and first boot. Everything you need to join the openclaw hardware revolution.
This guide walks you through building a complete DIY ClawBox build using a Raspberry Pi 5 — the most popular and well-supported platform. The total cost is approximately $150-200, and the build takes about 2-3 hours from unboxing to your first AI conversation.
No soldering is required. No special tools beyond a small Phillips screwdriver. If you've ever assembled IKEA furniture or installed an operating system, you have all the skills you need. The desktop AI openclaw experience is designed to be accessible to everyone.
Unbox your Raspberry Pi 5 and place it on a clean, anti-static surface. Before mounting anything, take a moment to familiarize yourself with the board layout — note the GPIO header on the right, the PCIe connector on the left, the USB-C power port, and the MIPI display/camera connectors. The Pi 5 is a remarkably dense piece of engineering with everything you need on a single board.
If you're using a openclaw AI assistant kit from a community vendor, the board may come pre-installed in a case with the cooler attached. If so, skip to step 4.
The Pi 5's CPU can generate significant heat during sustained AI workloads. The official active cooler combines a heatsink with a temperature-controlled fan, keeping your board cool and preventing thermal throttling that would slow down inference.
Remove the protective film from the thermal pad on the cooler's heatsink. Align the cooler with the mounting holes on the Pi 5 board — the fan connector should be oriented toward the fan header near the USB ports. Press down gently but firmly to seat the thermal pad against the CPU. Secure with the two plastic push-pins through the mounting holes. Connect the fan's 3-pin JST connector to the fan header on the Pi 5 board — the connector is keyed and only fits one way.
AI models are large — often 4-15 GB each — and they need fast storage. The Pi 5's PCIe 2.0 x1 interface provides approximately 500 MB/s throughput, which is more than adequate for loading models quickly.
First, connect the NVMe HAT to the Pi 5's PCIe FPC connector — the flat ribbon cable connector on the left side of the board. The cable has a specific orientation; the blue tab typically faces up on the Pi 5 end. Carefully insert the ribbon cable into the connector, ensuring it's fully seated, then close the locking tab.
Next, install your NVMe SSD into the M.2 slot on the HAT board. M.2 2230 or 2242 form factor SSDs fit directly; 2280 SSDs may need a mounting adapter. Insert the SSD at a slight angle (about 30 degrees) into the M.2 connector, then press down and secure with the included screw. Mount the HAT to the Pi 5 using the provided standoffs and screws.
For the ClawBox openclaw setup, we recommend Raspberry Pi OS (64-bit, Bookworm) or Ubuntu Server 24.04 LTS. Both provide excellent ARM64 support.
On your other computer, download and install the Raspberry Pi Imager. Insert your microSD card, launch the Imager, and select Raspberry Pi OS Lite (64-bit) as the operating system. Choose your microSD card as the storage target.
Before writing, click the gear icon (Advanced Options) and configure: hostname (e.g., "clawbox"), enable SSH with password authentication, set a strong password, and optionally configure WiFi. These settings will be applied during the first boot, saving you from connecting a keyboard and monitor.
Write the image to the SD card, then insert it into your Pi 5's microSD slot. Connect Ethernet and power. The Pi will boot, and you can SSH in within about 60 seconds.
Once booted, we'll migrate the OS to the NVMe SSD for better performance:
With the hardware configured and OS running, it's time to make your ClawBox look like the polished device it is. Aluminum cases provide excellent passive cooling and a professional appearance. 3D-printed cases offer more customization and often include mounting points for displays, cameras, or additional peripherals.
Carefully place the Pi 5 assembly (with cooler and NVMe HAT attached) into the case. Most cases for the Pi 5 with NVMe HAT have specific cutouts — align the ports with the openings in the case. Secure with the case screws, taking care not to overtighten. The case should feel solid with no rattling or loose components.
If your case includes a fan, connect it to the appropriate GPIO pins (typically 5V and ground) or directly to the fan header. Configure fan control in the OS to keep noise minimal during idle periods.
With the hardware ready, install the OpenClaw agent framework. SSH into your ClawBox and run:
The installation takes about 10-15 minutes depending on your internet speed. Once complete, you can access the web interface at http://clawbox.local:3000 or configure additional interfaces like Telegram or Discord bots. See our Software Setup Guide for detailed configuration options.
Your openclaw private AI hardware is ready! Open the web interface, type your first message, and experience the magic of a fully local AI assistant. Try asking it to summarize a document, write some code, or just have a conversation. Everything is processed on your device — no data leaves, no one is listening, and the only limit is your hardware (and imagination).
For voice interaction, connect your USB microphone and speaker, then enable the voice pipeline in the settings. For home automation, install the Home Assistant plugin. For development, connect your IDE to the local API. Your ClawBox is now the hub of your personal AI ecosystem — and it answers only to you.
Congratulations! You've built a DIY AI box that rivals commercial cloud AI assistants in capability while exceeding them in privacy, independence, and personalization. Welcome to the community.
Ensure the EEPROM is updated (sudo rpi-eeprom-update -a) and boot order is set to NVMe first. Some NVMe SSDs require external power — check compatibility lists on the DIY electronics forums for known-working models.
First load of a model from Ollama downloads it — subsequent loads are fast. Check that your NVMe is detected (lsblk). If using SD card only, model loading will be significantly slower.
Try a smaller model: ollama pull llama3.2:1b. Ensure the active cooler is running (vcgencmd measure_temp should stay under 80°C during inference). Close other applications to free RAM.
Check ALSA device recognition: arecord -l and aplay -l. USB audio devices sometimes need to be reconnected after boot. Verify the voice pipeline is enabled in OpenClaw settings.
For more help, visit our OpenClaw setup guides or join the community Discord.
| Hardware assembly | 30 min |
| OS installation | 20 min |
| NVMe migration | 15 min |
| Case assembly | 15 min |
| Software installation | 15 min |
| Model download | 10-30 min |
| Configuration | 15 min |
| Total | ~2-3 hours |