yessglory17/generative-mui

Generative MUI is a library that lets an AI chatbot or agent build actual, working user interface elements instead of only returning plain text. Normally when a chatbot wants to show a chart, a form, or a table, developers either hand write a matching component for every possible case, or let the AI output raw HTML, which brings security risks and inconsistent styling. This project takes a middle path: the AI model outputs structured data describing the interface using a protocol called A2UI, and the library turns that description into real Material UI components that match the visual theme already used in your app. Instead of sending a full nested webpage, the AI describes the interface as a flat list of pieces referenced by an id, with the actual data, like a typed in name or a form value, kept separate and linked by reference. This separation means a text field can update live as the user types, and a submit button can automatically disable itself if a required field is empty, all without a fresh reply from the AI. The library processes this stream of instructions through a few clear steps: parsing and validating the messages, applying them to an internal state in a predictable way, and finally rendering the current state as real interface components. Because the AI's output could arrive incomplete or malformed, the library is built to handle that gracefully. Interface pieces that have not fully arrived yet show a loading skeleton, broken or looping structures fall back to a safe placeholder instead of crashing, and any interface component type the library does not recognize is simply skipped rather than executed as code. Style properties coming from the AI are also blocked from directly reaching the real components, closing off one route for style based attacks. The project is split into two packages: a framework neutral core that handles parsing and state with no dependency on React, and a separate adapter that renders that state as Material UI components in a React app. It also ships ready made tool definitions so an AI model using common agent frameworks can produce this structured interface data directly.