FAQ

LlamaBox is a high-performance, local AI inference appliance. It's a dedicated piece of hardware that runs AI models on your network — not the cloud. You connect your own AI agents (OpenClaw, Hermes, custom scripts) to the LAN endpoint, and they communicate with the model running on LlamaBox. Your data never leaves your building. No subscriptions. No cloud dependency.

ChatGPT sends your prompts to OpenAI's servers. LlamaBox processes every request locally on your network. ChatGPT requires an internet connection and a monthly subscription. LlamaBox works offline and is a one-time purchase. ChatGPT only works with OpenAI's assistant. LlamaBox works with any OpenAI-compatible agent — yours or ours. Most importantly: ChatGPT sends your data to a cloud vendor. LlamaBox keeps your data on your LAN. Period.

No. LlamaBox is designed to be "set it and forget it." It boots up, broadcasts its API endpoint on the LAN, and you're ready to connect your agent. If you're not technical, we can help configure your agent for you (professional services at $200/hr). If you are technical, you have full control over every aspect of the setup.

LlamaBox keeps working. It's designed to run 100% offline after initial setup. No internet? No problem. Your AI runs locally, on your network, regardless of your internet connection. This is a core feature, not a fallback.

Absolutely — that's the point. LlamaBox provides an OpenAI-compatible API endpoint (http://llamaxbox:8080/v1). Any agent that speaks that format works: OpenClaw, Hermes, custom scripts, web UIs, CLI tools. You bring your own agent. We provide the model. You're not locked into our ecosystem.

New Qwen versions drop? We push updates to your appliance. No hardware changes needed. Just a model file swap — the same hardware runs the new model. Model updates are free for all LlamaBox customers. If you have a support contract, we handle the update remotely. If you don't, we provide the model file and instructions.

Every tier runs Qwen3.6 models at optimal quantization for the GPU. Starter runs Qwen3.6-35B (fits in 24GB). Pro runs Qwen3.6-70B at 4-bit quantization (fits in 48GB). Ultra runs Qwen3.6-110B at 4-bit quantization (fits in 96GB). You can swap models anytime — we provide model files and support the swap.

Yes! The RTX GPU is visible through the tempered glass side panels. The GPU is the "engine of privacy" — it's what processes your AI requests locally. The glass case design is intentional: it proves the AI is running on-site, not in the cloud. Premium matte black branding on the front bezel. Furniture-ready design that looks great on a desk or in an office.

Yes. All tiers are fully customizable. Need more RAM? More storage? A different GPU? We'll build it. You can also swap the GPU later — same OS image, same API endpoint, just a hardware upgrade. The model file changes, not the appliance.

Every LlamaBox ships with a backup USB drive. You can restore from backup in minutes. If the hardware fails completely, we'll ship a replacement appliance. If you have a support contract, we handle remote diagnostics and troubleshooting. All components are standard, off-the-shelf parts — easy to replace.

LlamaBox is for any business that handles sensitive data and needs AI capabilities. Primary targets: law firms (client data confidentiality), CPAs/financial advisors (regulatory compliance), medical offices (HIPAA), small businesses (privacy + cost savings), and anyone who values data privacy over convenience.

Yes. LlamaBox ships in professional packaging with foam. Hand delivery for the Louisville area. For remote locations, we use PC shipping boxes with custom foam cutouts. The appliance is built to survive transit. Just remember: LlamaBox is LAN-only. It won't broadcast on the internet — it's designed for a local network.

LlamaBox is designed with maximum privacy in mind. The firewall defaults to DENY ALL. Only ports 22 (SSH) and 8080 (inference) are open, both LAN-only. No internet exposure. No inbound access. Your data literally never leaves your building. This isn't a feature — it's the design principle.