Savindu Thathsara is embarking on a project to create a lightweight Linux operating system with an offline AI assistant for normal laptops. The goal is for users to control their laptops by voice commands that are processed locally, without sending data to the cloud. This project is being documented as a public build log, with the initial installation yet to occur. The target hardware is a standard Lenovo ThinkPad with 8GB RAM and integrated graphics, ensuring the system performs well on common machines. The focus is on delivering few, reliable features over a wide range of poorly performing ones.The offline AI encompasses speech recognition, intent selection, and voice replies all running on the device for privacy. A small, local Language Model acts as a fallback for complex phrasing. While network access for tasks like YouTube and updates remains, core AI decisions are local. The assistant will have a fixed, predictable set of skills rather than trying to interact with everything on screen.Initial skills include playing YouTube videos, opening apps, adjusting system settings, finding files, web searches, setting reminders, and taking screenshots. Sensitive actions will always require user confirmation. The AI is explicitly prevented from inventing shell commands, ensuring honesty about its capabilities. A key feature is local memory, allowing the assistant to learn user habits and personalize interactions over time.Performance metrics are crucial, including boot time, idle RAM usage, RAM during assistant operation, wake word to action completion speed, and skill accuracy. The target idle desktop RAM is 400-600MB, with the always-on components using 150-250MB. The system is initially being developed for English on a minimal Debian with LXQt, with future support for other languages planned. The project emphasizes transparency by showcasing constraints like ordinary hardware and visible user data.