Hey, this is a known issue that several users have been reporting recently. There’s a main thread tracking it here: Taking longer than expected
To figure out what’s causing the slowness in your case, I’ll need a few things:
A Request ID from one of the slow requests. Click the three-dot menu at the top right of the chat and select Copy Request ID
Run network diagnostics: Cursor Settings > Network > Run Diagnostics, and paste the results here
In the meantime, try this:
Open App Settings with Ctrl+,, search for HTTP/2, and enable Disable HTTP/2. This fixes a lot of these slowness issues, especially on certain networks.
Close Cursor, go to %APPDATA%\Cursor\User\globalStorage, and check the size of state.vscdb. If it’s several GB, delete it along with any .backup or .corrupted files, then restart.
Let me know how it goes, or paste the diagnostics and request ID and I’ll dig deeper.
I was wondering what to do about this file, my state.vscdb is huge and the whole software is ultra slow, takes 3 minutes to close even. when agent is running i can’t do anything else but type (means i can type in the editor window etc but hover over response takes minutes, opening new files takes minutes etc).
It didn’t start to get this bad until 2.5 and worse in 2.6 now.
I am really worried about deleting that file because it contains all the chatlogs, code tracking lines, etc.. what ACTUALLY happens if i delete it? My file is about 10GB now
@Evan_Lee glad turning off HTTP/2 fixed it. That confirms it was a network layer issue. Since it’s fast again, I’ll mark this as resolved. If it comes back, ping me with a new request ID.
@f00z fair point. Here’s the deal with state.vscdb:
This compacts the database without breaking anything. Quick heads up, if the file is huge, some users say it doesn’t shrink much, so results can vary.
If VACUUM doesn’t help, at least back up the file before deleting it, so you can restore it if needed.
About the general slowness like taking 3 minutes to close and UI freezes while the agent runs. A 10 GB state file can absolutely cause that. The team knows about the unbounded growth issue and it’s being tracked.
Can you also share your Cursor version and try turning off HTTP/2 like above? That might help with agent timeouts separately from the state.vscdb issue.
Aye, vacuuming does nothing.. the lag doesn’t show up anywhere in the ‘live’ cpu view in process explorer, it’s barely using anything there like 1-2%, but it shows up here:
Why can’t you guys give us some way to clean up the database file? this file has literally been the bane of my existence since cursor was made and i’ve been using it since 0.something, a long time ago, and this file is just absurd. Putting that much in a sqlite db the way it’s in there (i looked at it) is absurd.
Give us a way to remove the older code tracking / timeline/ logs /etc. Or at THE LEAST make it archive that to other files that it can then compress.
I’m afraid to delete it. I have deleted it before but i need to reinstall the whole program at that point and then i do lose my chats (although i noticed now that the program is storing summarized transcripts of the chats in the workspace which is good).
Auto mode is consistently very slow for me.
If I manually select any available model, responses are fast.
As soon as I switch back to Auto, the slowdown returns immediately.
Steps to Reproduce
Open a new chat
Send a short request in Auto mode → very slow
Switch to manually selected models → fast
Switch back to Auto → slow again
Expected Behavior
Auto mode should be at least comparable to manually selected models and should not be dramatically slower.
The issue happens specifically in Auto mode.
If I manually select any available model, responses are fast.
As soon as I switch back to Auto, the slowdown returns immediately.
Happens on linux also today. I’ve purchased Pro a couple of days ago. This is the first time I am trying to work, after the purchase, but effectively I cannot. I get very slow responses. Cursor always says: “it takes longer than normal..“
