v2 Interactive Followup Questions - Best Practice Question

[dhruv-sanio]

In Vanna v0.x, followup questions were generated via endpoint and rendered as
clickable buttons by the UI. We're migrating to v2 using the LegacyVannaAdapter and
want to maintain this UX.

We created a GenerateFollowupQuestionsTool that works, but the questions appear as
text in the Agent response rather than interactive buttons.

Questions:

1. What's the recommended v2 pattern for interactive UI elements from Tool outputs?
2. Are there v2 UiComponent types for interactive question lists?
3. Can custom Flask endpoints work alongside VannaFlaskServer?

Would appreciate guidance on maintaining this v0.x feature in the new architecture!

[zainhoda]

It's possible to customize this. You're able to emit a button group with buttons for follow-up questions. By default those buttons will send the action back to the chat interface.

Those buttons are also optionally able to trigger a deterministic workflow as well. Have a look at the DefaultWorkflow to see how the initial UI does "Show Memories"

Let me give some thought to how best to handle this in general and I'll try to publish an example in the docs. I think if you want to have the LLM make a determination about what to suggest as follow-up questions, one possibility is to give it a tool like suggest_questions(list[str]) that will return the buttons as UI components.