[Enhancement]: Change the design of the thoughts recursion so that it doesn't fill up the chat area and handle cases gracefully when the thought recursion rans out

[Odrec]

What features would you like to see added?

When some models need to handle complex processes the thought recursion fills up the chat area making it difficult to read and ugly for users. (See image)

Can the design be changed so that several thoughts are integrated into a single expander or another solution that avoids the clogged chat area?

Also when a model reaches the recursion limit it just throws an error. Is there a way to handle this more gracefully like imposing a stop condition on the last cycle?

More details

The same as above.

Which components are impacted by your request?

UI

Pictures

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

This often due to the model itself or parameters on the hosting provider. A small context window could also be the cause

[Odrec]

We are using gpt-oss through vLLM and it's a 4 bit variant so it doesn't surprise me too much that it gets confused.

But we saw similar behavior in the Anthropic models through VertexAI with the entire context. It doesn't reach the error but, for some complex tasks while using the web search, it fills the chat area with a bunch of "Thoughts" and it doesn't look good or user friendly unfortunately.