44.11 Composer Agent constantly stalling / stuck on generating

same, extremely infuriating

While experiencing the same issues noted here, I began to wonder if Cursor has a “real” beta tester program or if this is that. I imagine direct interaction with the developers on features and bugs would yield far greater and faster results. I have done this a few times with companies that make audio plugins and it helped speed up the issue resolution. Just a thought, though not entirely sure anyone from Curser is here or how they check in on this platform.

I am also working on extending an extension for dynamic .cursorrules file generation that has helped me significantly with combating all of the issues with AI’s wide adoption and unavailability, as many have articulated here.

A good approach still today is to accept that this is still “new,” not just with Cusror but with everything, [arguably], and ALL of them are pretty “buggy” at the moment… I have to do this with every AI I use today, so I have accepted that as a “current state” and pondered solutions to that rather than remaining frustrated by something outside my control.

The point is, with what I have created for myself, it constantly updates my .cursorrules file so that I can point any AI, even if I want to switch to a different one midstream and provide very accurate definitions of what I am doing within about 30mins to 1hour of accuracy. While I agree that this should just be “fixed” it’s not, so workarounds will be a reality for some time.

Version: 0.44.11
VSCode Version: 1.93.1
Commit: fe574d0820377383143b2ea26aa6ae28b3425220
Date: 2025-01-03T07:59:06.361Z
Electron: 30.5.1
Chromium: 124.0.6367.243
Node.js: 20.16.0
V8: 12.4.254.20-electron.0
OS: Darwin arm64 24.2.0

3 Likes

Can we get some verification that this is being worked on? I have a push to setup Cursor at my company but this level of agent unresponsiveness and failing is about to torpedo that whole initiative and I really don’t want to get stuck with Windsmurf or other solutions.

any updates on addressing this? Composer agent responsiveness seems to be getting worse over time, I’m now having composer agent getting completely stalled out and stuck on generating 5-10 mins into each new session.

I love the idea of Cursor and have had some benefit from using it at points but for me at least it’s become virtually unusable and i am stuck trying to wrap up a project…Anyone have any good alternatives to recommend?

Same here. I guess hopefully they get a fix i just started using it and I thought it was just me. But, i love the program other than the hangups. How do i get the previous version that is stable?

good question - not sure how to revert tbh

Hey all – really sorry to hear you’ve been running into this too. I can confirm the dev team are indeed hard at work on exactly this issue right now.

I’m not a Cursor employee, but I have been experiencing the same issue (I feel your pain!) and I’ve had chance to help with reproducing it this afternoon. It’s a surprisingly elusive edge case, but now that there are some more examples I think they’re already making good progress on it.

From what I can tell, it’s been more complex than expected because it seems there’s more than one trigger. At least one cause of it has already been eliminated, but clearly that’s not the whole picture yet.

If you’re seeing it reproducibly, your input may actually be really valuable:

  • Temporarily switch off ‘Privacy’ mode if you have it on, otherwise they get almost no information about your requests at all
  • Enter a prompt and observe a ‘stuck’ generation attempt
  • In the editor, press Cmd+Shift+P (mac) or Ctrl+Shift+P (PC) for the command drop-down, and type Report AI Action. From there you can get the request ID for your prompt attempt
  • Share the request ID that failed and which LLM model you were using

One of the challenges at scale is if only 1% of users are seeing a particular issue, that actually might be a really huge deal, and yet it can be really hard to reproduce without Report AI Action reports. The system’s designed to allow for the “Privacy” mode, which is great, but it means if that’s switched on they’ll see virtually no details about your request. Following the steps above can make a world of difference to their debugging.

Hang in there, and hopefully we’ll see a fix for this very shortly. If I hear more, I’ll be sure to let you know.

3 Likes

I might be wrong, but from what I can tell it’s not exclusive to 44.11. I’m entirely speculating here, but I wonder if it’s more a service-side issue than the IDE itself, which means a downgrade wouldn’t be much help. Much of the magic in figuring out how to get the LLMs (Claude, OpenAI GPT) to play nicely happens in the stage between the IDE and the model.

The one thing I can say for sure though is it’s a top priority and they’re actively working to resolve this, even as we speak.

2 Likes

I think I might be onto something here that may help some people like it helped me.
I was also struggling with generating issue and it is really annoying. I tried many things but just half an hour ago I had an idea to go through the files that were recently changed, this expandable window above the chat/composer input:


I noticed there was a file that was deleted, I brought it back clicking on it and after that chat was working again and is still working. You can try that or check if there are any files that might cause the issue even if you haven’t deleted anything, maybe you moved something. Don’t know if it helps but I wanted to share it, let me know if it works for someone :slight_smile:

2 Likes

Thanks for the responses. I will be patient. Seems to be working fine right now.

Thanks for the responses. I will be patient. Seems to be working fine right now.

just pushed a speculative fix for this - please let me know if anyone is still seeing this issue! request ids and videos are very helpful!

3 Likes

i’m now getting non stop “Connection Failed” messages. this has now surpassed the endless generating. have started new Composer sessions probably about 30 times today as a result and pretty much unable to get anything done

I am stuck on generating. As soon as I submit my message in Composer, I see the three horizontal dots animating, the ‘Cancel generation’ button, and this appear in console:

Error: Image selected in conversation was not found on disk
    at d (workbench.desktop.main.js:3518:65690)
    at async workbench.desktop.main.js:3518:66866
    at async workbench.desktop.main.js:3800:8951
    at async Promise.all (index 0)
    at async Promise.all (index 4)
    at async workbench.desktop.main.js:3800:9628
    at async Promise.all (index 4)
    at async ie.populateConversationWithExtraContext (workbench.desktop.main.js:3800:8146)
    at async Kt.submitChatMaybeAbortCurrent (workbench.desktop.main.js:4952:67360)
    at async Object.hi [as onSubmit] (workbench.desktop.main.js:5070:34030)
console.error @ workbench.desktop.main.js:3542
(anonymous) @ workbench.desktop.main.js:3800
await in (anonymous) (async)
(anonymous) @ workbench.desktop.main.js:3800
populateConversationWithExtraContext @ workbench.desktop.main.js:3800
(anonymous) @ workbench.desktop.main.js:1287
m @ workbench.desktop.main.js:1287
f.value @ workbench.desktop.main.js:1287
submitChatMaybeAbortCurrent @ workbench.desktop.main.js:4952
await in submitChatMaybeAbortCurrent (async)
hi @ workbench.desktop.main.js:5070
await in hi (async)
onClick @ workbench.desktop.main.js:4694
(anonymous) @ workbench.desktop.main.js:2598
1 Like

I then scrolled up in the chat, found a previous message with an image attached, deleted it, and that got it responding again, but then mid-response it stalled with this repeating in the console:

workbench.desktop.main.js:465 [6ed] potential listener LEAK detected, having 175 listeners already. MOST frequent listener (1):

||check|@|workbench.desktop.main.js:465|
| --- | --- | --- | --- |
||q|@|workbench.desktop.main.js:468|
||J|@|workbench.desktop.main.js:1302|
||o|@|workbench.desktop.main.js:1168|
||createInstance|@|workbench.desktop.main.js:1168|
||I|@|workbench.desktop.main.js:1315|
||createModel|@|workbench.desktop.main.js:1316|
||(anonymous)|@|workbench.desktop.main.js:4832|
||He|@|workbench.desktop.main.js:298|
||xe|@|workbench.desktop.main.js:298|
||Je|@|workbench.desktop.main.js:298|
||Se|@|workbench.desktop.main.js:298|
||(anonymous)|@|workbench.desktop.main.js:298|
||Te|@|workbench.desktop.main.js:298|
||Ee|@|workbench.desktop.main.js:298|
||Te|@|workbench.desktop.main.js:298|
||Oe|@|workbench.desktop.main.js:298|
||Zt|@|workbench.desktop.main.js:298|
||ft|@|workbench.desktop.main.js:5070|
||onClick|@|workbench.desktop.main.js:5070|
||(anonymous)|@|workbench.desktop.main.js:4709|

workbench.desktop.main.js:465 Error at g.create (workbench.desktop.main.js:465:58394) at b.q [as onDidChange] (workbench.desktop.main.js:468:1100) at new J (workbench.desktop.main.js:1302:8515) at h.o (workbench.desktop.main.js:1168:1609) at h.createInstance (workbench.desktop.main.js:1168:1101) at u.I (workbench.desktop.main.js:1315:1976) at u.createModel (workbench.desktop.main.js:1316:963) at Object.fn (workbench.desktop.main.js:4832:1007) at He (workbench.desktop.main.js:298:26587) at xe (workbench.desktop.main.js:298:26354) at Je (workbench.desktop.main.js:298:27977) at Se (workbench.desktop.main.js:298:29228) at workbench.desktop.main.js:298:28842 at Te (workbench.desktop.main.js:298:28157) at Ee (workbench.desktop.main.js:298:28835) at Te (workbench.desktop.main.js:298:28169) at Oe (workbench.desktop.main.js:298:26009) at Zt (workbench.desktop.main.js:298:20173) at ft (workbench.desktop.main.js:5070:73022) at Object.onClick (workbench.desktop.main.js:5070:74858) at HTMLDivElement.<anonymous> (workbench.desktop.main.js:4709:12112)

||check|@|workbench.desktop.main.js:465|
| --- | --- | --- | --- |
||q|@|workbench.desktop.main.js:468|
||J|@|workbench.desktop.main.js:1302|
||o|@|workbench.desktop.main.js:1168|
||createInstance|@|workbench.desktop.main.js:1168|
||I|@|workbench.desktop.main.js:1315|
||createModel|@|workbench.desktop.main.js:1316|
||(anonymous)|@|workbench.desktop.main.js:4832|
||He|@|workbench.desktop.main.js:298|
||xe|@|workbench.desktop.main.js:298|
||Je|@|workbench.desktop.main.js:298|
||Se|@|workbench.desktop.main.js:298|
||(anonymous)|@|workbench.desktop.main.js:298|
||Te|@|workbench.desktop.main.js:298|
||Ee|@|workbench.desktop.main.js:298|
||Te|@|workbench.desktop.main.js:298|
||Oe|@|workbench.desktop.main.js:298|
||Zt|@|workbench.desktop.main.js:298|
||ft|@|workbench.desktop.main.js:5070|
||onClick|@|workbench.desktop.main.js:5070|
||(anonymous)|@|workbench.desktop.main.js:4709|

I ended up deleting all of the image attachments in the chat and that allowed me to get another response, but then it ended up cancelling with this appearing in the console:

workbench.desktop.main.js:3542 [composer] ToolFormer: error in toolWrappedStream undefined
workbench.desktop.main.js:3542 ConnectError: [resource_exhausted] Error
    at F.$endAiConnectTransportReportError (workbench.desktop.main.js:3755:12992)
    at T.S (workbench.desktop.main.js:2950:14846)
    at T.Q (workbench.desktop.main.js:2950:14612)
    at T.M (workbench.desktop.main.js:2950:13645)
    at T.L (workbench.desktop.main.js:2950:12480)
    at b.value (workbench.desktop.main.js:2950:11175)
    at $.B (workbench.desktop.main.js:470:732)
    at $.fire (workbench.desktop.main.js:470:949)
    at s.fire (workbench.desktop.main.js:1153:14824)
    at G.onmessage (workbench.desktop.main.js:3000:8229)
workbench.desktop.main.js:3542 [composer] Error in AI response: ConnectError: [resource_exhausted] Error
    at F.$endAiConnectTransportReportError (workbench.desktop.main.js:3755:12992)
    at T.S (workbench.desktop.main.js:2950:14846)
    at T.Q (workbench.desktop.main.js:2950:14612)
    at T.M (workbench.desktop.main.js:2950:13645)
    at T.L (workbench.desktop.main.js:2950:12480)
    at b.value (workbench.desktop.main.js:2950:11175)
    at $.B (workbench.desktop.main.js:470:732)
    at $.fire (workbench.desktop.main.js:470:949)
    at s.fire (workbench.desktop.main.js:1153:14824)
    at G.onmessage (workbench.desktop.main.js:3000:8229)
workbench.desktop.main.js:3542 [composer] Failed to get complete AI response


Hope this helps @rishabhy

Guys this is still happening to me. Yesterday i needed to start several more composer chats because it eventually stalled and got stuck in “generating…”

My guess is this is related with context size in the composer chat, because it only happens after it gets big.

1 Like

Same here, Cursor times out every 1-2 hours and I have to start with a new window every time. This is getting worse lately.

So sorry to hear this is ongoing. If you’re finding this is happening consistently, then you’re in the perfect position to help resolve it. Please do consider taking a moment to:

  • Temporarily switch off ‘Privacy’ mode if you have it on (to stop the request being redacted)
  • Enter a prompt and observe a ‘stuck’ generation attempt
  • In the editor, press Cmd+Shift+P (mac) or Ctrl+Shift+P (PC) for the command drop-down, and type Report AI Action. From there you can get the request ID for your prompt attempt
  • Share the request ID that failed along with which LLM model you were using

If even one or two people affected were able to do this, it would be much appreciated. I think it could be a significant help in narrowing down the issue.

1 Like

This seems to be working much better today!

1 Like