Agent hangs on "Taking longer than expected" after 3.21.4 update (Windows)

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Product: Cursor IDE / Agent
Version: 3.21.4 (user setup)
Commit: 6294c6f4f2868bf530f04ea320bd44ab7ecb3490
Date: 2026-09-16T01:42:06.337Z
OS: Windows_NT x64 10.0.26100
Layout: IDE
Build: Stable / Default
Summary
After updating to 3.21.4 today, Agent became mostly unusable. Network Diagnostics all pass. Changing HTTP Compatibility Mode did not fix it. Some turns return quickly; most hang forever on “Taking longer than expected”. Cannot work.
What we already tried

  • Deleted oversized %APPDATA%\Cursor\User\globalStorage\state.vscdb (~3 GB) plus -wal/-shm — did not help
  • HTTP Compatibility Mode changed (HTTP/1.1 / HTTP/1.0)
  • Network Diagnostics: OK
  • New chats, specific model (not Auto), Stop/resend

Steps to Reproduce

  1. Open Agent chat
  2. Send a prompt (also happens with a short “hello”)
  3. UI shows “Taking longer than expected” and usually never recovers

Expected Behavior

Expected: Agent streams a response
Actual: Intermittent hang; majority of turns stuck

Operating System

Windows 10/11

Version Information

Version: 3.21.4 (user setup)
VS Code Extension API: 1.128.0
Commit: 6294c6f4f2868bf530f04ea320bd44ab7ecb3490
Date: 2026-09-16T01:42:06.337Z
Layout: IDE
Build Type: Stable
Release Track: Default
Electron: 42.10.0
Chromium: 148.0.7778.280
Node.js: 24.18.1
V8: 14.8.178.38-electron.0
xterm.js: 6.1.0-beta.291
OS: Windows_NT x64 10.0.26100

For AI issues: add Request ID with privacy disabled

c188642c-45f7-49e6-9fb2-764fb6cfc37d

Does this stop you from using Cursor

Yes - Cursor is unusable

Hey @Sergio_Herbas,

The Request ID you shared never left your machine (same for the other hung turns), while the ones that went through completed fine. So it’s getting stuck locally before anything is sent - that’s why Network Diagnostics passes and HTTP Compatibility Mode does nothing (you can set that back to default).

A couple of quick tests would pin it down:

  1. Fully quit Cursor (end any leftover Cursor processes in Task Manager) and relaunch.
  2. Open a brand-new empty folder (File → Open Folder), start a new Agent chat, send “hello” - does it reply or hang?
  3. Back in your normal project, run “Terminal: Kill All Terminals” (Ctrl+Shift+P) and clear any @-attachment from the chat box, then try “hello” in a new chat.

Next time one hangs, run “Developer: Capture and Send Debugging Data” (Ctrl+Shift+P) - with Privacy Mode on it saves a local zip. Attach it here (or email [email protected] referencing this thread) with the Request ID and rough time; that’s what lets us pin down where it’s stuck.

This isn’t on your end, and we’ve let the team know. No need to delete state.vscdb again - that only clears chat history and isn’t related here.

Hi,

The suggested steps were followed. For point two, the agent responded almost immediately. I switched to my usual project, ran “Terminal: Kill All Terminals,” performed a test, and the agent responded correctly. This is the request ID: e8a8b4a3-bfb0-49a0-9ed0-9919b7d62e41. I opened a new chat and requested a task that the agent normally handled without issue, but it hung. This is the request ID: c99a8721-7a14-47cd-98c6-7248ef5c1f70. The zip with the logs is attached.

The time was between 8:50 and 9:10 AM Eastern Time.

Regards

(attachments)

debugging-data-1789650444423.zip (5.13 MB)

Hi,

The suggested steps were followed. For point two, the agent responded almost immediately. I switched to my usual project, ran “Terminal: Kill All Terminals,” performed a test, and the agent responded correctly. This is the request ID: e8a8b4a3-bfb0-49a0-9ed0-9919b7d62e41. I opened a new chat and requested a task that the agent normally handled without issue, but it hung. This is the request ID: c99a8721-7a14-47cd-98c6-7248ef5c1f70. The zip with the logs is attached.

The time was between 8:50 and 9:10 AM Eastern Time.

Regards

Hi,

The suggested steps were followed. For point two, the agent responded almost immediately. I switched to my usual project, ran “Terminal: Kill All Terminals,” performed a test, and the agent responded correctly. This is the request ID: e8a8b4a3-bfb0-49a0-9ed0-9919b7d62e41. I opened a new chat and requested a task that the agent normally handled without issue, but it hung. This is the request ID: c99a8721-7a14-47cd-98c6-7248ef5c1f70. The zip with the logs is attached.

The time was between 8:50 and 9:10 AM Eastern Time.

Regards

The issue has been fixed for me with version 3.21.13+.

Thanks for that capture, @Sergio_Herbas. Your two request IDs line up: the one that worked connected out normally, and the one that hung never left your machine, so the turn is getting stuck locally before anything is sent (which is why Network Diagnostics passes and the HTTP toggle does nothing). We’ve passed it to the team with your logs. This isn’t anything on your end, and it’s an issue we’re tracking.

On “fixed in 3.21.13+”: worth updating to the latest to see if it settles for you, but I won’t over-promise, since we’re still seeing this on the newest builds too. In the meantime, when a turn hangs, open a new chat or run Reload Window and resend (it’s intermittent per turn), and keep to one chat with fewer attachments, which seems to hang less. If it keeps hanging on the latest build, run Developer: Capture and Send Debugging Data on a fresh hang and drop the new zip, request ID, and rough time here.