Make the Renderer Prev Response or Position Aware with UI persistence

[AbhinRustagi]

Would be interested in the scope discussion for this:

Context: Working on a fluid dashboard, with a floating text box, where the dashboard layout/information card/graphics change with prompts. While that is completely possible with the current renderer, it would be helpful if we could persist components across sequential responses from the model. I couldn't find a way to do this right now with the current renderer.

The way I have done it right now -> Pass the card position in the grid as a prop (in this case, would be provided by the model). Persist the pinned components' position, and insert the pinned component in the model response at its pinned position before sending it to the renderer.

What I'm designing for is to give users the ability to handcraft their own KPI dashboards. These pinned components can be stored in their raw OpenUI Lang structure, and glue them with appropriate React Query hooks for fresher data.

Should this be handled outside the renderer? I reckon the renderer can be updated to be prev response/component-position aware with UI persistence.

Quick Prototype of what I'm building with current approach

[AbhinRustagi]

Another hacky fix that I've tried: Add multiple renderers to the parent, and manually split the model response and distribute across the multiple renderer instances.

[rabisg]

at a high level your approach looks great till now. I couple of other ways off the top of my head:

• Don't generate 1 dashboard but rather generate 5 widgets via 5 LLM calls and then you arrange it yourself. is that what you were doing?
• another way can be you parse the response manually. allow only one child type within root call widget and then have multiple widget types as union within the parent widget type then manually parse the response and do operations on it - not ideal because this is unnecessary manual work on top of LLMs

ps: We are working on a proper way to support this