I don’t get it - how is this useful?
2 Likes
I got this feature as well with no documentation or anything. It seems to me like it’s not something actually finished yet that was accidentally released in the wild*. It seems like it’s intended to let you run multiple agent runs (most usefully with different models) and then either compare and pick an optimal one, or somehow merge them to create an even more ideal solution to the problem.
Currently, it makes git branches/worktrees out of the different runs but there’s not logic or UI element to pick which one you want to use, to compare them or even get rid of the worktree bits when they inevitably break your normal git operations. And as I said, there is no documentation about this feature anywhere on the internet
(* Wild meaning “Early Access” so yeah it’s not wrong to see experimental or undocumented features, but this one seems either much less finished or much less intuitively designed from a UX perspective than what I am expecting from Cursor.)
2 Likes
The feature you are describing is complete. If you received access, you have opted into “early access” from settings → beta. Because of this, you will get “early access” to features coming out soon.
Documentation has been live: Worktrees / Local Parallel Agents | Cursor Docs
2 Likes
Indeed, yesterday or the day before when I first saw this feature and tried it, this google search did not have such a result. Though, even reading this it’s not exactly clear what’s going on:
I can see the tabs and click them but there is no “keep” button which the documentation is referring to. The only thing somewhat similar to “keep” is the “Apply All” text just above the text entry box but this is not normally something I interact with at all since all changes made by the agent are in fact already applied automatically as they are done - this “Apply All” has always just meant to approve the highlighted changes in open files.
2 Likes
Ah, I think we need to change the docs to say “Apply All,” which brings the content to your local branch
1 Like
The docs are clear, thanks, however why is there an option to run the same prompt multiple times in parallel using the same model? This seems redundant, as the worktrees will presumably be highly similar.
Models can have very different approaches to challenging or visual tasks - I use it when I’d like to see 2 different approaches on a hard task
2 Likes
They should be calling 3+ ai ensembles. They don’t seem to know what an ensemble is. Thought they are ml experts? No just young guns. No one wants to use a single ai for coding because they all have different strengths and stochastics. Please give me my ensemble I am happy to pay 3 times to get it right once.
[Admin - Condor: Redacted due to violation of Forum Rules]
@Luke_Poga Please stay civil and avoid name calling. You may have misunderstood things. In Cursor you can run x agents in parallel with different models.
2 Likes
The whole thing seems really buggy. Tried stupid-quick model1 and deepreasoner model2.
Apply (or Keep or whatever the name is today) interfered/confused model2 still running. Should be isolated to be able to test what model1 did.
I also don’t wanna see stash popups, should be stuff you handle under the hood. Nobody wants to battle with git. “Failed to apply to current branch”, thank you!
Also much stuff missing, and constant connection issues, e.g. chat history not loading, cheetah slooooow, models not finishing the work, etc.
This all feels a lot like vibe slop. I just want a working tool.
Back to 1.7 I guess.
1 Like
I’m a bit confused also. There’s a lot of workflow stuff that isn’t really explained with this.
1 Like