Cursor version 2.5.22 is unusable: endless waits for AI responses, the editor often crashes and automatically restarts Windows (I haven’t seen anything like this in 20 years). What’s going on? Why are you releasing versions with all these problems? Don’t tell me to go back to previous versions, that’s absurd because the previous versions are full of problems too!
Steps to Reproduce
Chat with AI
Expected Behavior
That the AI responds in a reasonable amount of time and that the IDE does not crash.
Hey, we need to check a few things since you actually have two different issues happening at the same time.
Windows restart crashes are very unusual. We need to understand what exactly is happening:
Does Cursor crash and restart by itself, or does Windows reboot?
Are you running long chat sessions before the crash? If yes, this could be a known Windows renderer out-of-memory issue. The process can hit the ~4 GB memory limit and crash. You can monitor this via Ctrl+Shift+P > Developer: Open Process Explorer.
If the editor specifically freezes right when the agent asks for confirmation, try this:
In Cursor: Settings > General > Notifications, turn off System Notifications
In Windows: Settings > Notifications, find Cursor and turn it off
Slow AI replies: please run the network diagnostics: Cursor Settings > Network > Run Diagnostics and paste the results here. Also try this: Settings Ctrl + , > search for “HTTP/2” > enable “Disable HTTP/2”. This helps on some networks.
Can you share the crash logs too? Check %APPDATA%\Cursor\logs for any recent crash files.
I think it is indeed an “out-of-memory issue.” I waited a long time for the AI to respond (between 5 and 10 minutes, I don’t remember exactly), after which I asked the AI if it had frozen. When it didn’t respond, I clicked on Stop (the button to stop the AI process). At that moment, Cursor froze, Windows gave me a message that it needed to restart and restarted automatically. I couldn’t do anything.
After restarting Windows, I reopened Cursor and the AI continued to respond very slowly. I closed Cursor and reopened it after about an hour, tried to work with it, and some AIs responded within normal times (Opus 4.6), and I was able to finish a task that I hadn’t finished before. I am attaching the logs, but the crash happened hours ago, and I don’t know if the logs are the right ones.
———————–
Cursor Network Diagnostic Results
DNS: Success
Logs:
[2026-02-23T18:27:45.219Z] Host: api2.cursor.sh
[2026-02-23T18:27:45.219Z] Servers: 192.168.1.1,fe80::321f:48ff:fe11:5bac
[2026-02-23T18:27:45.219Z] Resolved to 44.216.223.4 in 26ms
[2026-02-23T18:27:45.251Z] Resolved to 44.216.223.4 in 5ms
[2026-02-23T18:27:45.253Z] Resolved to 44.216.223.4 in 1ms
[2026-02-23T18:27:45.345Z] Resolved to 44.216.223.4 in 17ms
[2026-02-23T18:27:45.357Z] Host: api2.cursor.sh
[2026-02-23T18:27:45.357Z] Servers: system
[2026-02-23T18:27:45.357Z] Resolved to 44.195.108.117, 54.227.121.228, 23.20.44.121, 54.166.225.188, 100.52.59.113, 3.89.51.114, 18.208.84.238, 44.216.223.4 in 5ms
[2026-02-23T18:27:45.358Z] Resolved to 44.195.108.117, 54.227.121.228, 23.20.44.121, 54.166.225.188, 100.52.59.113, 3.89.51.114, 18.208.84.238, 44.216.223.4 in 0ms
[2026-02-23T18:27:45.359Z] Resolved to 44.195.108.117, 54.227.121.228, 23.20.44.121, 54.166.225.188, 100.52.59.113, 3.89.51.114, 18.208.84.238, 44.216.223.4 in 0ms
[2026-02-23T18:27:45.360Z] Resolved to 44.195.108.117, 54.227.121.228, 23.20.44.121, 54.166.225.188, 100.52.59.113, 3.89.51.114, 18.208.84.238, 44.216.223.4 in 1ms
[2026-02-23T18:27:45.360Z] Result: true
Authentication UI: Success
Logs:
[2026-02-23T18:27:45.176Z] DNS lookup: authenticator.cursor.sh
[2026-02-23T18:27:45.251Z] Resolved authenticator.cursor.sh to 104.18.35.233 in 34ms
Cursor Tab: Success
Logs:
[2026-02-23T18:27:45.176Z] DNS lookup: api3.cursor.sh
[2026-02-23T18:27:45.219Z] Resolved api3.cursor.sh to 104.18.19.125 in 14ms
Agent Endpoint: Success
Logs:
[2026-02-23T18:27:45.176Z] DNS lookup: agent.api5.cursor.sh
[2026-02-23T18:27:45.248Z] Resolved agent.api5.cursor.sh to 100.30.166.181 in 28ms
Codebase Indexing: Success
Logs:
[2026-02-23T18:27:45.176Z] DNS lookup: repo42.cursor.sh
[2026-02-23T18:27:45.249Z] Resolved repo42.cursor.sh to 52.204.241.114 in 28ms
Thanks for the logs, they’re really informative. They show two clear issues:
SQLite concurrency bug: Your logs are flooded with AgentAnalyticsOperationsMainService: SQLITE_ERROR: cannot start a transaction within a transaction, hundreds of times between 16:12 and 16:18. This is a known issue where the analytics service gets stuck in a loop of failed DB writes, which eats CPU and memory.
Bloated local database: The logs show DB size: 5860462592b, that’s about 5,8 GB for a local storage file that should be small. One key alone, aiCodeTrackingLines, is 3,1 MB. This causes about a 7 second delay just to close the DB, and it almost certainly contributes to freezes and rising memory use.
Here’s what I’d suggest:
Fully close Cursor
Go to %APPDATA%\Cursor\User\globalStorage
Find the folder that contains state.vscdb and check its size. If it’s a few GB, delete state.vscdb and also any state.vscdb.backup or state.vscdb.corrupted.* files
Restart Cursor (you’ll lose some local state, but this should fix the bloat)
For the future:
Fully restart Cursor (quit and relaunch) every 1 to 2 hours during long sessions
Start new chats instead of endlessly continuing very long ones
The team is aware of the SQLite transaction issue. Your logs are very helpful as extra evidence. Let me know how it goes after clearing the database.
also having cursor crash issues since updating to 2.5.25
am on a mac book pro M4+pro
seems to happen when agent running, in my case two chats at the same time, never seen the crashes before.