Continuation of timeout/interruption of chat

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Sorry I got busy with something else and didn’t make the time cut-off of this older thread.

It works sometimes, but get’s hung up periodically. But it’s doing something new, I’m assuming as a result of an attempted bug fix.
After going for a bit, it goes back to the prompt almost as if I never hit enter except it has listed out thoughts and explorations.
I keep retrying after it stops, it rarely starts working again.

It no longer hangs on ‘planning next moves’ or gives me the ‘taking longer than expected’.

Steps to Reproduce

Make any request big or small

Operating System

MacOS

Version Information

Version: 3.5.38
VSCode Version: 1.105.1
Commit: 009bb5a3600dd98fe1c1f25798f767f686e14750
Date: 2026-05-26T21:32:06.537Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Darwin arm64 25.3.0

For AI issues: which model did you use?

Company forced auto. Not sure how to tell.

For AI issues: add Request ID with privacy disabled

Did a new agent tab: 2a4561ff-8402-4533-b2e4-8b0660b6ff98
The original, long running tab: 4866a009-b80b-4368-8f56-ed58c9f80015

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

Following up from your earlier thread where @deanrie asked for more details. Thanks for the version info and request IDs here.

This is something we’ve been seeing across multiple users, and your description of the behavior changing (no longer showing “Taking longer than expected” but instead silently returning to the prompt) is consistent with reports we’re tracking. Our team is actively investigating.

To help narrow down what’s happening on your end, could you share two things?

  1. Network diagnostics: Cursor Settings > Network > Run Diagnostics, then click “Copy diagnostics” and paste the results here. This checks whether your network setup (proxy, VPN, firewall) is affecting the streaming connection.

  2. A per-message request ID from a failed chat: When the chat “goes back to the prompt” after showing thoughts, click the three-dot menu (…) on the last message that appeared, then Copy Request ID. The IDs you shared are conversation-level IDs, which we can work with, but a per-message ID gives us more precise server-side data.

Also, as a quick workaround, try adding this to your settings: open Settings (Cmd+,), search for HTTP/2, and enable Disable HTTP/2. Some users have seen improvement with HTTP/1.1 for streaming reliability.

Let me know how it goes.

Thanks for your quick reply. I appreciate the work you’re doing to get this fixed. In the settings HTTP/1.1 is selected.

Message Id: bd103142-511b-4878-93ef-9905834b9822

Cursor Network Diagnostic Results.txt (4.9 KB)

The network diagnostics look healthy. Looking at the message ID you shared, the request doesn’t appear to have reached our servers. This is consistent with a client-side issue we’ve identified where the chat can get into a stuck state between turns - specifically when the client thinks a previous generation is still active (when it’s actually already finished), new submissions get blocked silently.

This explains the behavior you’re describing: thoughts/explorations appear from a prior turn, but your new message never actually fires.

Workarounds until the fix ships:

  1. When it happens: Open a new chat tab (Cmd+T in the chat panel) rather than retrying in the same one. The stuck state is per-tab.

  2. If it recurs frequently in a session: Reload the window - Cmd+Shift+P > type “Reload Window” > Enter. This clears all stale state.

  3. Pattern to avoid: If a response seems to stall or partially complete, give it a moment before sending a new message in the same tab. The deadlock is more likely to trigger when a new message is sent right after a turn that ended ambiguously.

Our team has identified the root cause and is tracking a fix. I’ll follow up when it ships.

Thank you so much for your reply. I’ll keep this in mind and await your new version.