FAQ

facts. is a self-contained AI assistant that lives entirely on a USB flash drive. Plug it into any Mac, Windows, or Linux computer, launch one file (double-click on Mac and Windows, one script on Linux), and you're chatting with a fully functional AI — no internet, no installation, no accounts, and no trace left on the machine when you're done.

It runs Qwen 3 (Abliterated), a 4-billion parameter language model with its content restrictions removed, powered by the open-source llama.cpp inference engine. Everything runs locally from the drive and your computer's RAM.

Anything you'd use a conversational AI for — writing, coding, research, brainstorming, analysis, creative projects, learning — except it runs completely offline and privately. No one can see your prompts or responses. There are no content filters deciding what you're allowed to ask.

Some popular use cases: writing code without uploading proprietary work to the cloud, researching sensitive topics without creating a search history, drafting documents on air-gapped machines, and having a portable AI assistant that works anywhere — planes, remote locations, networks you don't trust.

Three fundamental differences. Privacy: ChatGPT sends every message to OpenAI's servers. facts. never connects to the internet — your conversations exist only in RAM and are destroyed when you close the session. Censorship: ChatGPT has extensive content filters. facts. uses an abliterated model — it answers directly without moralizing or refusing. Ownership: ChatGPT is a subscription service that can change terms, raise prices, or shut down. facts. is a physical device you own forever. One payment, lifetime use.

Abliteration is a fine-tuning technique that removes the refusal mechanisms baked into most AI models. Standard models are trained to decline certain prompts — even harmless or academic ones. Abliterated models answer questions directly, objectively, and without adding moral disclaimers. The Qwen 3 model on your drive has been abliterated to maximize utility and eliminate unnecessary gatekeeping.

Three steps. Step 1: Plug the drive into any Mac, Windows, or Linux computer. Step 2: Open the drive and double-click MacLaunch.command (Mac), WindowsLaunch.bat (Windows), or run LinuxLaunch.sh (Linux). Step 3: A terminal window opens with AI ready to chat. Type your question and press Enter. That's it. No installation, no configuration, no accounts.

When you're done, press Ctrl+C to exit. Unplug the drive. Zero trace remains on the host machine.

Mac: Apple Silicon (M1 or later) or Intel, with macOS. Intel Macs run in CPU mode (slower, fully supported). Minimum 8GB RAM — the launcher auto-detects your hardware and loads the appropriate model. 16GB+ RAM loads the high-resolution Q8 model for best quality. Under 16GB loads the efficient Q4 model.

Windows: Windows 10 or 11 (64-bit) on an Intel or AMD processor with at least 8GB RAM. The launcher auto-detects your hardware and selects the best model. GPU acceleration is automatic on supported systems — no drivers or setup needed. Windows on ARM is not supported — that means Snapdragon-powered machines, including Copilot+ PCs and most newer Surface laptops. If your Windows machine has a Snapdragon or Qualcomm chip, the drive won't run correctly on it.

Linux: Any modern 64-bit distro (Ubuntu, Fedora, Arch, Debian, Mint, etc.). 8GB+ RAM. GPU acceleration works on NVIDIA, AMD, and Intel graphics via Vulkan when drivers are installed.

USB 3.0 port recommended (USB-C with data adapter also works). No internet connection is ever required — not during setup, not during use, not ever.

Both are the same core product: a full offline AI on a flash drive, zero-log, no internet ever. facts. ($29.99) runs Qwen 3 4B and supports Apple Silicon and Intel Macs, Windows, and Linux.

facts. v1.5 ($49.99) upgrades the model to Qwen 3.5 4B (sharper reasoning, better at code), reformats the drive itself with 128KB allocation units so the AI loads faster, runs a first-launch benchmark that picks your machine's fastest hardware path, sizes the AI's memory to your RAM automatically, and ships with a custom drive icon.

If you already own the original, it keeps working forever — v1.5 is an upgrade, not a replacement.

Yes, this is completely normal. Windows Defender flags the AI engine as an "unknown publisher" because it's an open-source binary that isn't signed with a Microsoft certificate. This happens with many open-source tools.

To proceed: click "More Info" on the blue warning screen, then click "Run Anyway." This only happens once — Windows remembers your choice for future launches. llama.cpp is a widely-used, MIT-licensed open-source project trusted by millions of developers worldwide.

Linux: Yes! facts. now ships with a LinuxLaunch.sh script that works on any modern 64-bit Linux distro — Ubuntu, Fedora, Arch, Debian, Mint, Pop!_OS, you name it. Just plug in, open the script, and go. GPU acceleration works on NVIDIA, AMD, and Intel graphics via Vulkan when drivers are installed; otherwise it runs on CPU.

Chromebook: Not supported at this time. Most Chromebooks don't have enough RAM or the right architecture to run local AI models.

RAM is everything. The AI loads a full language model into your computer's memory — the more RAM available, the smarter and faster it runs. Windows alone idles at 6-11GB depending on your setup. Add a browser and you're eating into the AI's breathing room.

If the AI seems dumber than expected: close your browser and any other heavy apps before launching. You want at least 4GB of free RAM when the drive boots. The launcher shows your available RAM on startup — if it's tight, close some apps and relaunch.

16GB+ RAM with other apps closed = best experience. 8GB = works, but close everything first. Under 8GB = won't run.

Still having trouble? Click the wrench icon in the bottom-right corner — our LLM Engineer can diagnose your exact setup and write you a custom fix on the spot.

Every time you launch facts., the script executes what we call a "Nuclear" wipe protocol. Before the AI even boots, it force-deletes all previous session files: .llama_history, llama.chat.history, and main.session — from both the drive and the host machine. When you exit, no conversation logs, no cache files, and no session data remain anywhere.

The AI model runs in your computer's RAM. When you close the terminal, that RAM is released. There is no persistent storage of your conversations, period.

Never. The inference engine (llama.cpp) runs 100% locally. There are no API calls, no telemetry, no update checks, no analytics, no phone-home mechanisms of any kind. The drive is air-gapped by design. You can verify this yourself by running it with your WiFi off or monitoring network traffic with a tool like Wireshark — you'll see exactly zero outbound connections.

The Faraday bag is an electromagnetic shield that prevents any wireless signals from reaching or leaving whatever is inside it. It's an extra layer of physical security for your drive — in transit and in storage. Any order can ship sealed inside one: check the "Ship sealed in a Faraday bag" option when adding a drive to your cart, or grab a bag on its own (Small $5.99 / Large $8.99) from the store. Once you have the drive, the bag doubles as shielded storage for it or any small device you want off the air.

We source manufacturer-recertified drives from a major flash manufacturer. These are unused units that were repackaged to meet compliance regulations — they've never stored a single file before ours. We open each drive to hand-install the complete AI software suite, test it, and verify it works before shipping.

Think of "refurbished" as a supply chain label, not a quality indicator. The drives are in excellent condition with zero prior use. We chose this sourcing model to keep the price at $29.99 — roughly one-third of what competitors charge for similar products.

30-day replacement guarantee from OSE. If your drive arrives defective or stops working within 30 days, we replace it — no questions asked. All support is handled directly by our team. Contact us and we'll sort it out.

Please note: the drive's original manufacturer does not provide support for our software bundle. All support inquiries should be directed to OSE.

Yes, all the way down. The inference engine (llama.cpp) is MIT-licensed. The AI model (Qwen 3) uses a permissive open-source license. Every license is included on the drive. Every component is verifiable. We also publish the full parts list and build process — you can build one yourself at home for free if you have the technical knowledge.

Our launch scripts, troubleshooting guides, and update files are all public on GitHub: github.com/WEAREOSE/facts

We charge $29.99 for the convenience of a pre-loaded, tested, ready-to-run device with everything configured. You're paying for our time, testing, and packaging — not for the software.

Absolutely. That's the point. You'll need: a USB flash drive (32GB minimum), the llama.cpp binary for your OS, a GGUF-format model file (we use Qwen 3 4B Abliterated), and our launch scripts. Everything we ship on our drives — launchers, guides, fix files — is public on our GitHub: github.com/WEAREOSE/facts. If you want to skip the setup and just have a working drive in your hand tomorrow, that's what the $29.99 is for.

All updates, fixes, and the latest launch scripts are published on our GitHub: github.com/WEAREOSE/facts

If you're having trouble with your drive, check the Releases page for drop-in fix files, or message our support team and we'll point you to the right one.

We ship from the United States to United States addresses only. Every order is assembled, loaded, and QA-tested by hand, so it takes 2 to 7 days before yours goes out. After that, standard USPS delivery usually runs another 3-7 business days. Most orders arrive within 1 to 3 weeks total, and you'll get a tracking number by email as soon as it ships.

Yes — orders start at $10 (before shipping). We're a small operation and every order is packed and shipped by hand, so tiny single-accessory orders end up costing us more than they bring in. Add a second bag or pair it with a drive and you're set.

We accept all major credit cards (Visa, Mastercard, Amex, Discover), Apple Pay, and Google Pay — all processed securely through Stripe.

If your drive arrives defective, we'll replace it within 30 days. Because each drive is individually loaded with software and personally tested, we handle refund requests on a case-by-case basis. Contact our support team and we'll work something out — we're real people, not a corporate helpdesk.