fix(virtual-quota): display active model in UI for the frontend

[ivanarifin]

Context

Help user to know the active model while using the virtual quota fallback with correct model name and correct context window

Implementation

When the backend switches the model, it now sends out a "model has changed" signal by emitting event.
The main application logic catches this signal and immediately tells the user interface to refresh itself.
The user interface then updates the display to show the name of the new, currently active model.
This will also keep the backend and the frontend active model in sync

Screenshots

How to Test

Use virtual quota fallback as a provider

Get in Touch

ivanarifin

[changeset-bot]

🦋 Changeset detected

Latest commit: a3d9802

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
kilo-code	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[ivanarifin]

Hey matt @mcowger, would you mind testing this PR? tagging you here because you are the creator of the feature

[mcowger]

Will give it a shot tomorrow. Thanks for the ping.

[mcowger]

I took a try with this. It does appear to switch the displayed information for the active model in the task timeline view, but only for the first change.

E.g., If I configured:

1. qwen3-coder-plus (128kl)
2. gemini-2.5-flash (1M)
3. gpt-5-codex (400k)
4. qwen3-coder (262K)

The context timeline at the top shfts from 128K to 1M as expected on the first transition. However, subsequent transitions occur (in that the request is sent to the correct profile), but the state of the context tracker doesn't change and remains at 1M context.

[ivanarifin]

I took a try with this. It does appear to switch the displayed information for the active model in the task timeline view, but only for the first change.

E.g., If I configured:

1. qwen3-coder-plus (128kl)

2. gemini-2.5-flash (1M)

3. gpt-5-codex (400k)

4. qwen3-coder (262K)

The context timeline at the top shfts from 128K to 1M as expected on the first transition. However, subsequent transitions occur (in that the request is sent to the correct profile), but the state of the context tracker doesn't change and remains at 1M context.

@mcowger
I can't seem to reproduce your issue
Here is my testing with 4 profiles and all RPM set to 1

Screen.Recording.2025-10-15.at.06.24.37.mov

[mcowger]

Very strange.

I'll give it another try tonight and see if I can grab some logs.

[mcowger]

I was wrong - I think I may have been testing in the wrong window.

Confirmed both the context window tracker, as well as the quick picker by the chat window change as expected. This also appears to fix the context condensing bug.

Well done!

The only thing I think you need here is the //kilocode_change tags.

@chrarnoldus this works well for me!

[ivanarifin]

well I was actually working on the context condensing bug at first, then I thought, why not also fix the UI

[ivanarifin]

@chrarnoldus any chance this will be merged?