Claude has artifacts, which appears to be a very flexible way to bring interactive components to the UI. It uses XML tags to mark artifacts in the UI, which can then be interpreted by external renderers like a browser. This could be used for showing graphs, react code rendering etc.

Somebody posted the Sonnet 3.5 artifacts instructions on twitter which might be a good starting point for implementation details, but you might want to consider a more general solution. In theory, artifacts should work with any model. It's just code within a specific XML tag that gets interpreted a certain way by the application. Having a UI that allows workflows like v0.dev or Claude artifacts would be amazing, where I can switch between viewing code and seeing the rendered result and then iterating over it.

There are already some existing client that have done it, so would be nice to have this in here.