Why call it an operating system?
Khan OS began as a question about interfaces. Most AI assistants live inside a chat box, even when they can use tools, understand voice, and retrieve live information. I wanted to explore what the experience would feel like if intelligence were the main interaction layer instead of a feature added to a conventional app. The result is a React and TypeScript workspace with multimodal input, real-time voice interaction, live data grounding, and tool use for actions such as alarms and tasks.
Separating conversation from action
The most important architectural boundary was between understanding a request and executing it. A user saying remind me tomorrow contains intent, time, and an action, but those pieces should be validated before a tool runs. I designed the flow so the assistant interprets the request, produces a structured action, confirms ambiguous details, and then invokes the appropriate capability. This makes the system easier to extend and reduces the chance that conversational text is treated as an unchecked command.
Making the interface feel alive
The visual system uses glass-like layers and a neural sphere animation to communicate listening, thinking, and acting states. The animation is not only decoration. It gives the user feedback when audio is being captured, when a response is streaming, and when a tool is running. I kept status transitions explicit because voice interfaces can feel broken when they go silent. Good motion should reduce uncertainty, not compete with the content.
The lesson behind the polish
Khan OS looks like a design-heavy project, but the difficult work was state management. Voice sessions, partial responses, tool results, errors, and live data can all arrive asynchronously. A polished interface quickly falls apart if those states are not modeled carefully. The build taught me to treat latency, cancellation, confirmation, and recovery as first-class product features. The best AI experience is not the one that appears magical. It is the one that stays understandable when the magic fails.