FAQ — Frequently Asked Questions

Frequently Asked Questions

What exactly am I buying?

VaultAI is a high-speed NVMe SSD pre-loaded with optimized AI models and a professional desktop application. You plug it into your Mac or PC via USB-C, launch the app, and immediately have a private AI assistant. Everything runs locally on your computer — no cloud, no accounts, no internet required.

How is this different from free local AI tools?

Free local AI tools exist, and they work — if you're comfortable at a command line, picking models, matching them to your hardware, and wiring up everything around them. What you're paying for with VaultAI is the finished product: a curated lineup of models chosen and tuned for real jobs, a polished app with chat history, projects and a prompt library, private search across your own documents, image generation, and storage that travels between machines. Zero setup. Plug in and work.

Do I need internet to use VaultAI?

No. VaultAI works 100% offline. The only feature that requires internet is the optional web search (powered by DuckDuckGo). All AI models, document processing, and image generation work without any internet connection.

What if my computer doesn't meet the requirements?

VaultAI runs on your computer's CPU and RAM — the SSD provides the models and application. Check your system requirements:

Core: 16GB RAM minimum (most Macs and PCs from 2020+)
Pro: 32GB RAM minimum, 64GB recommended
Ultra: 64GB RAM minimum, 128GB recommended

If your computer doesn't have enough RAM, the larger models will run slowly or not at all. VaultAI automatically detects your hardware and recommends the best models for your system. We offer a 14-day return policy if VaultAI doesn't work with your setup.

Where do you ship?

Across the United States, and internationally to Canada, the UK, Ireland, and most of Western Europe, plus Australia, New Zealand, Japan, South Korea, Singapore, Hong Kong, Malaysia, Israel and the UAE. Enter your address at checkout to see live rates. International orders may be subject to customs duties or import taxes set by your own country — those are not included in the price.

Can I add my own models?

Yes. Every drive ships with free space left over, and you can add more models from inside the app.

How do I get updates?

Software updates are available for download from our website. You replace the application file on the SSD — your data, models, and settings are untouched in separate directories.

What's your return policy?

14-day money-back guarantee. If VaultAI doesn't work with your system or doesn't meet your expectations, email us for a return authorization and send it back for a full refund of the purchase price. There's no restocking fee. Return shipping is yours to arrange — unless the device arrived defective, in which case we reimburse it up to $25.

Is this HIPAA compliant?

VaultAI is designed for privacy-sensitive workflows. All data processing happens locally on your device — nothing is transmitted to any server. However, HIPAA compliance requires organizational certification beyond just the technology. VaultAI provides the technical foundation for compliant workflows, but your organization should consult with a compliance officer.

Does it work on both Mac and Windows?

Yes. Each VaultAI SSD includes both macOS and Windows versions of the application and launcher scripts. Plug into either platform and go.

How fast is the AI?

Speed depends on your computer's hardware. On a MacBook Pro with Apple Silicon:

Smaller models (3B-8B): 20-40 tokens/second — very responsive
Medium models (27B-32B): 10-20 tokens/second — good for most tasks
Large models (70B+): 5-10 tokens/second — slower but more capable

VaultAI automatically detects your hardware and shows performance estimates for each model, so you can choose the right balance of speed and capability.

What models are included?

Core (21 models): GLM-4 9B, Gemma 4 E4B and E2B, Mistral 7B, Qwen3 4B, Ministral 3 3B, Zephyr 7B Lite and Gemma 3 4B for chat; Devstral Small Lite and Ornith 1.0 9B for code; Moondream 1.8B and LLaVA 7B for images; MedGemma 1.5 4B and MedGemma 4B Lite for health; five uncensored specialty models; plus Z-Image-Turbo for image generation.

Pro (35 models): everything in Core, plus GPT-OSS 20B and its uncensored twin, DeepSeek-R1 32B, Gemma 4 26B, Mistral Small 3.2 24B and GLM-4 32B for chat; Codestral 22B, Qwen3-Coder 30B and Ornith 1.0 35B for code; Qwen3-VL 8B and 32B and MiniCPM-V 8B for images; the full MedGemma 27B; and a second image engine, FLUX.2 Klein 4B.

Ultra (46 models): everything in Pro, plus the frontier lineup — Qwen3-235B-A22B, GPT-OSS 120B, DeepSeek-R1 70B Uncensored and IBM Granite 4 Small-H 32B with a 1-million-word memory — along with Qwen3.6 35B-A3B, Qwen3 32B Uncensored, Hermes 3 8B Uncensored, Ministral 3 8B, Qwen3.5 9B and the IBM Granite 4 Micro and Tiny-H models.