fluxOS

Inspiration

Modern software is static. Users are forced to adapt their unique workflows to rigid menus, fixed dashboards, and fragmented applications. We have millions of apps, yet none of them fit us perfectly. We asked ourselves: "What if the interface could adapt to us?"

What it does

fluxOS is a "Liquid UI" operating system that builds itself in real-time. It has no fixed buttons or permanent menus. Instead, it uses the Gemini 3 API to interpret your intent via voice or text and dynamically constructs a React-based interface tailored to that specific moment.

If you say "I need to track expenses for my Tokyo trip," fluxOS dissolves its current state and morphs into a currency-converted expense table with a receipt scanner. If you then say, "Show me where I spent the most money," it instantly transforms that table into a 3D interactive heatmap. It’s not an app; it’s an interface that generates the right tool, right when you need it.

How we built it

We built fluxOS using Next.js and React for the frontend, styled with Tailwind CSS and Shadcn UI for a clean, modern aesthetic.

The core "brain" is Gemini 3. We leverage:

1. Structured Outputs: We don't just ask Gemini for text; we force it to output complex JSON schemas that dictate the exact layout, component hierarchy, and data binding of the UI.
2. Function Calling: Gemini determines which real-world data sources (like exchange rates or map data) are needed before generating the UI.
3. Generative Component System: Instead of writing raw HTML, Gemini acts as an orchestrator, selecting from a library of pre-built, modular React components to assemble the interface in milliseconds.

Challenges we ran into

The biggest challenge was "hallucinating" a UI that is actually functional. Early versions would generate beautiful but broken interfaces. We solved this by creating a strict "Component Map"—a contract between the AI and the frontend that ensures every JSON object returned by Gemini corresponds to a valid, interactive React component.

Accomplishments that we're proud of

We are incredibly proud of the "morphing" effect. Seeing the interface dissolve and rebuild itself based on a simple natural language command feels like magic. It proves that the future of UI isn't about better menus; it's about no menus.

What we learned

We learned that Gemini 3 is surprisingly good at visual reasoning. It understands that a request for "trends" implies a line chart, while a request for "locations" implies a map, without us explicitly programming those rules.

What's next for fluxOS

We plan to integrate Deep Thinking for complex workflows (e.g., "Build a CRM for my dog walking business"), allowing Gemini to plan a database structure before rendering the UI. We also want to add Multimodal Input, so you can show fluxOS a physical object, and it will generate the interface to interact with it.

Built With

• gemini-3-api`
• next.js`
• react`
• shadcn-ui`
• tailwind-css`
• typescript`