add thinking_delta field to AgentStream events to expose Ollama's reasoning (feat: [19270])

[Stoic-Angel]

Description

• Add thinking_delta field to AgentStream class with default empty string
• Update FunctionAgent, CodeActAgent, and ReActAgent to pass thinking_delta from ChatResponse
  Fixes [Feature Request]: Expose Ollama "thinking" in FunctionAgent's AgentStream #19270

Type of Change

Please delete options that are not relevant.

• New feature (non-breaking change which adds functionality)

How Has This Been Tested?

Your pull-request will likely not be merged unless it is covered by some form of impactful unit testing.

• I added new unit tests to cover this change

Suggested Checklist:

• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• I have added Google Colab support for the newly added notebooks.
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I ran uv run make format; uv run make lint to appease the lint gods

[AstraBert]

Looks good, just two minor comments :)

[AstraBert]

I would use default_factorty = str. Non-blocking, tho

[AstraBert]

Can we add some more comprehensive test that mocks the execution of an agent and streams the thinking delta? Just to make sure that everything is captured correctly

[Stoic-Angel]

on a side note: @AstraBert do you think we could possibly modularise the CI tests, i.e. focusing on smaller sub-packages for any particular change? just curious :)

[AstraBert]

For sub-packages like integrations, tests happen directly for those packages that have been changed, but for changes occurring in llama-index-core we want to make sure no unintended breaks are introduced, and the only way to do it is to always test for all core modules, even when the change is contained... Hope this explanation helps!

[AstraBert]

Looks good!

[Stoic-Angel]

Of course, makes sense. Thank you! :D

[iKotori96]

Perhaps using last_chat_response.message.additional_kwargs.get("thinking_delta", "") is a more reasonable choice.

[Stoic-Angel]

Valid question. In Ollama's ChatResponse implementation, the message param contains the FULL thinking text. When I fetch the thinking_delta however (which is not in message but in ChatResponse only), the response is returned chunk by chunk (since it's looping over r in response), thereby enabling streaming. I think that's better.