⚠️ No Changeset found

Latest commit: b981a33

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

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

This pull request includes a large number of changes across different areas of the codebase. To improve reviewability and maintain focus, it would be beneficial to split the localization changes into a separate pull request. This way, the core functionality changes related to the warning component can be reviewed independently from the localization updates. Please consider creating a separate pull request for the localization changes, which include updates to the i18n files for various languages.

Very cool! Will take a look today.

Would suggest playing around with a TaskHeader placement... :)

@sachasayan i get inspired from claude chat, i think they put near chat area to make UX better when your drag mouse into it, should we can add dismiss / close button like them ?

Okay, I'm fine with it where it is. Yes, I'd say let's add a dismiss/close button and let's ship. We can do a styling refinement pass later. :)

@mrubens

• for default value i think context window: 75%, total token (input tokens + output token): 5_500_000, cost: 3$
• for Should this setting be model specific: for total token may be we need model specific
• When it starts a new task, does that boomerang back when it’s done?: i think we don't need to boomerang back

@sachasayan after dismiss should we make it show up again when user meet 1-2 consecutive error ?

@mrubens

• for default value i think context window: 75%, total token (input tokens + output token): 5_500_000, cost: 3$
• for Should this setting be model specific: for total token may be we need model specific

Yeah, total tokens does seem like it would need to be model specific. Same thing with cost, I think.

• When it starts a new task, does that boomerang back when it’s done?: i think we don't need to boomerang back

Right now every new_task will boomerang back right?

Yeah, total tokens does seem like it would need to be model specific. Same thing with cost, I think.

should we move all into provider config or just total + cost ?

Right now every new_task will boomerang back right?

yeah, let make an follow up pr that can allow new_task had option to not return parent task ? and default will be return to parent ?

@mrubens update my pr, move config to per profile setting, can disable with switch or manual set to -1

Hey @samhvw8 I think we've chatted about this one and we're not quite sure about where the thresholds should be + it needs some UI polish. Those are the two big outstanding issues right now.

I think there’s also the issue that all new tasks are subtasks currently. Seems like we also need to figure out what to do about that.

Restated - what’s the action that we want the user to take when we warn them about the context? I think this could make a lot of sense if paired with a smol/compact action for instance, but would love to have a clear opinion on that.

Hey @samhvw8 I think we've chatted about this one and we're not quite sure about where the thresholds should be + it needs some UI polish. Those are the two big outstanding issues right now.

From @mrubens i move it to api profile

For Ui polish, can you help me and push commit direct to this pr ?

I will try to make it can dismis

Restated - what’s the action that we want the user to take when we warn them about the context? I think this could make a lot of sense if paired with a smol/compact action for instance, but would love to have a clear opinion on that.

Later we can show for them both compact and start new chat option ?

@sachasayan @cte @mrubens what do you think about this ?

https://www.loom.com/share/cbbeb3d93bfb4e918140da2663bf5601?sid=bbb221e6-a52b-4927-beea-cdea0d0b49b2

I'm curious to see how #3582 plays out. If it works well, we may not need this from a cost/context management perspective. Though the PR description focuses more on client-side performance...

@mrubens ok so i just close this pr for now

Thanks for the inspo here! We added some version of this in #3582, which you can enable with the autoCondenseContext experimental feature.
Some follow up features are tracked in these issues:

• Add manual trigger button for context condensing #3671
• Allow users to configure the autoCondenseContext threshold #3672
• Add a UI element for context condense operations #3673