Same nuclear-wipe, zero-trace experience. Two real engineering upgrades. The drive itself is custom-formatted at the filesystem level so the AI loads faster every time you plug in, and the launcher benchmarks your machine on first boot so it runs on whichever path is actually fastest.
8GB+ RAM required (16GB recommended) | Mac (Apple Silicon or Intel), Windows (Intel or AMD — no Snapdragon/ARM), Linux. Intel Macs run in CPU mode (slower, fully supported). The smart benchmark catches weak iGPUs that look fast on paper but are slower than the CPU in practice. Saves you from running on the wrong hardware path.
Filesystem. We don't just load files onto a stock drive. We low-level reformat every v1.5 with 128KB allocation units. The default exFAT format ships with 32KB clusters, sized for storing a mix of small files. Our drives store one 4GB AI model that gets read end-to-end every launch. Bigger clusters means the operating system makes fewer reads with less overhead per byte. The AI loads faster every time you plug in. And on machines tight on RAM, long conversations stay snappy instead of slowing down. Anyone can verify the format with one terminal command.
Model. Qwen 3.5 4B Abliterated, the next generation up from the Qwen 3 4B that ships on v1.0. Sharper reasoning, better at code.
Launcher. Runs a 30-second hidden benchmark on first launch, testing your GPU and CPU on the actual model. Locks in whichever runs faster. No more wasted cycles on weak iGPUs that load Vulkan but lose to a fast CPU in practice.
Branding. Custom volume icon shows in Finder and Explorer when mounted.