Everything you need for a smooth first week — setup steps, documentation and the answers to the questions new owners ask most.
Setting up the ClawBox AI hardware platform is designed to be measured in minutes, not weekends. The physical side is trivial: connect power, plug a network cable into your router, and wait for the status light to settle. There are no drivers to install and no hardware to assemble, because the unit arrives as a finished, tested appliance.
From there, open a browser on any device on the same network and navigate to the device's local address. The onboarding flow detects the hardware, walks you through an initial language-model download, and invites you to pair whichever messaging accounts you want to use. The whole process is guided and self-explanatory, and most people reach a working assistant in well under an hour.
Every channel shares the same memory and the same automations, so there is no separate configuration per platform. Pair once, and the assistant behaves identically everywhere it is reachable.
Once the assistant is online, the fastest way to see its value is to schedule a small automation. A daily morning summary, a price check on a product you are watching, or a monitor on a page that changes rarely are all good starting points. These run in the background on the device itself, so they continue whether or not you have the dashboard open.
Automations are configured through the dashboard in plain language rather than code. If you later want finer control, the REST API and the full Linux environment underneath give you every option — but the everyday path is deliberately non-technical.
The very first prompt can feel slow because the selected model is still loading into memory. After that warm-up, responses are noticeably faster, and the device keeps the model resident so subsequent requests are immediate. If you switch models, expect a brief load on the first use of the new one.
Yes. Local inference and local automation work fully offline. Only messaging integrations and software updates need a connection, so an air-gapped deployment is entirely possible for those who want the strictest isolation.
Your archive lives on the local NVMe drive and can be exported from the dashboard at any time. Export gives you a portable copy you can store anywhere — a second drive, a NAS or offline storage — so your assistant's memory is never hostage to a single device.
The documentation answers most questions, and the resource centre is updated as the software evolves. If you are stuck, the contact page puts you in touch with a human member of the team — there is no automated support maze.
Once you are comfortable with the basics, the three feature pages are the natural place to go deeper. Hardware and performance explains what the silicon can do, software and local AI covers the agent core and privacy model, and integrations and platforms walks through every channel in detail. The companion resource page on benchmarks and industry analysis is there when you want the broader context.
Once the assistant is online, the fastest way to build confidence is a small, concrete project. Each of these takes under an hour and demonstrates a different corner of the platform.
Ask the assistant to monitor a small set of pages — a news site, a forum thread, a product listing — and deliver a plain-language summary to your chosen channel each morning. This teaches you the scheduling system and shows how the browser automation and local summarisation models combine. Within a day or two you will be tuning exactly what it includes and how long the digest should be.
Point the device at a folder of documents — contracts, notes, saved articles — and let it index them. Then ask questions in natural language: "What did the vendor agree to about delivery dates?" or "Summarise everything I saved about the renovation." This is where the retrieval system shines, and it quickly becomes the feature people wonder how they lived without.
Use the local model as a drafting partner for the emails and documents you actually have to produce. Because everything stays on the device, you can paste sensitive drafts without hesitation and iterate on tone, length and structure until the result is right. The value here is not the raw model — it is that you can be candid with it.
Confirm the device and your computer are on the same network, and that the status light shows normal operation. If the address does not resolve, try the device's fallback local address printed in the quick-start guide, or connect over Ethernet temporarily while you investigate the Wi-Fi configuration.
The first request after a model switch is always slower while weights load into memory. Sustained slowness usually points to a heavier model than your workload needs — try the leaner default, or close a concurrent automation that is competing for the same GPU. The dashboard shows current load, so the cause is usually visible at a glance.
Most channel issues are token or permission expiry on the platform side, not the device. The dashboard's integration screen will flag exactly what needs re-authorising, and the fix is typically a fresh pairing in a minute or two.
You can reset the archive and configuration from the dashboard without affecting the underlying system, or perform a full factory reset that returns the device to its out-of-box state. Both are documented in the manual, and both are entirely under your control.
An always-on appliance should not require constant attention, and by design this one mostly does not. A small amount of periodic care, however, keeps it running well for years and protects the data you have accumulated on it. None of these tasks is hard, and together they take minutes a month.
The device checks for and installs updates automatically by default, and each release notes are published so you can see what changed. Updates are signed and delivered over the network, and they never touch your archive. If you prefer to stay on a known version, updates can be paused and applied manually instead.
Your assistant's memory is the most valuable thing on the device, and it deserves the same care as any other data you would regret losing. A monthly export to an external drive is cheap insurance. The export is a standard, portable format, so a backup made today can be read by the platform years from now — or inspected by you directly.
Over time the archive grows, and occasionally it is worth a quick review. The dashboard shows what is taking space — indexed documents, conversation history, downloaded models — and lets you prune any of it. Keeping one or two models rather than a collection also frees memory and storage for the tasks you actually run.
The only physical maintenance the device ever needs is occasional dusting of the vents to keep airflow unobstructed. There are no moving parts to lubricate and no battery to replace. If the device lives in a well-ventilated spot out of direct sunlight, it will quietly do its job for years without intervention.
Our team reads every message — and we are happy to walk you through setup personally.
Contact support