webui : store reasoning_content so it is sent back in subsequent requests

[aldehir]

Overview

The reasoning_content should be sent back in the assistant message on subsequent requests. It wasn't being saved in the history and therefore lost.

Additional information

For models that support interleaved thinking, it is important to send back the reasoning during agentic loops.

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: YES to find the gap

[aldehir]

Apologies, I did not realize the type check failed.

[allozaur]

Hey, @aldehir please re-run npm run build and we'll have this merged.

[aldehir]

@allozaur rebased and rebuilt.

[allozaur]

@ServeurpersoCom plz give 2nd approval and let's merge it :)

[julmb]

Isn't this model specific? The Gemma 4 model description says:

No Thinking Content in History: In multi-turn conversations, the historical model output should only include the final response. Thoughts from previous model turns must not be added before the next user turn begins.

Maybe there should be a flag embedded in the model metadata to specify this?

[ServeurpersoCom]

Isn't this model specific? The Gemma 4 model description says:

No Thinking Content in History: In multi-turn conversations, the historical model output should only include the final response. Thoughts from previous model turns must not be added before the next user turn begins.

Maybe there should be a flag embedded in the model metadata to specify this?

Good point, but this is already covered. On the server side, the chat template is the source of truth: if a model's template doesn't reference reasoning_content, the field is simply ignored. On the WebUI side, there's already a toggle: "Exclude reasoning from context" that lets the user strip it explicitly. So no extra flag needed.

[aldehir]

Isn't this model specific? The Gemma 4 model description says:

No Thinking Content in History: In multi-turn conversations, the historical model output should only include the final response. Thoughts from previous model turns must not be added before the next user turn begins.

Maybe there should be a flag embedded in the model metadata to specify this?

1. Reasoning should be retained for Gemma 4 models between tool calls. https://ai.google.dev/gemma/docs/core/prompt-formatting-gemma4

2. Models that don't need it will simply not include it in their templates. We delegate to the chat template, as @ServeurpersoCom mentioned.

[julmb]

I was not aware of the template stripping the reasoning as necessary, but that actually makes a lot of sense, thank you for explaining!