# Known limitations

- Vuka is based on Kali rolling. It inherits Kali's fast-moving package base and is not a substitute for a conservative long-term-support distribution.
- The ISO is larger than 4 GiB. Write it as an image with a tool that supports hybrid Linux ISOs; copying it onto a FAT32 filesystem is not sufficient.
- No LLM model weights are embedded. The first graphical login automatically downloads `llama3.2` (about 2 GB) and `llama3.2-vision` (about 7.8 GB) in the background. Until the text model finishes, Seven can start and speak but local reasoning is degraded. Live USB sessions without persistence must download the models again after reboot.
- Seven has unrestricted root and desktop authority by design. A compromised model, extension or prompt can therefore affect the whole machine. Vuka records root actions but does not sandbox Seven.
- Wine compatibility is application-specific. Microsoft Store/UWP applications, Windows kernel drivers, firmware flashers, many anti-cheat systems, current Microsoft 365 desktop and Adobe Creative Cloud remain unsupported or unreliable.
- Wine Mono is a compatibility implementation, not Microsoft's complete .NET Framework. Wine Gecko is a compatibility implementation, not current Microsoft Edge/WebView2.
- The VM acceptance suite cannot prove physical microphone quality, camera capture, Bluetooth, suspend/resume, robotics motion, HiDPI/dual-display behavior, GPU driver performance, RDP to a real Windows host, SMB interoperability or removable NTFS media. Those stay explicit physical-device gates.
- Seven's animated sprite/ASCII shell surface and voice runtime start by default. Visual animation and microphone enumeration can be proven in a VM, but audible speech recognition still depends on a working host microphone, VM audio-input routing and network access when Seven falls back to Google STT. The local Whisper package/model is not embedded.