Cursor AI responses are taking a long time to load. This happens even after basic troubleshooting, so I’d like help diagnosing whether this is a local app issue, account issue, network issue, or service-side latency.
Issue:
Responses are slow to start loading / slow to stream.
This happens in a totally empty Cursor window.
It also happens when using a hotspot instead of my normal network.
For any complicated task it is completely stuck on “planning” and never will even load
Hey, thanks for the detailed diagnostics. That really helps.
It looks like this isn’t a network or app bug. On Free, after you’ve used up the included usage, Auto still works, but requests go through a slower queued pool. Speed there depends on server load, so during peak hours responses can be very slow or get stuck on planning for harder tasks. These requests are marked as Free and aren’t billed.
That’s why troubleshooting didn’t help. If switching to a hotspot, HTTP/1.1, and an empty window didn’t change anything, it’s not your local setup, it’s the service-side queue. The diagnostics flagged proxy buffering, but that’s not the main factor here.
What you can do:
Wait for the monthly limit reset, then you’ll get fast speed again.
Or upgrade to Pro or Pro+ for faster responses and more usage.
Good news: the request you sent (Request ID df5d3282-c89b-4172-87ec-dd46120a2f76) did complete on our side. It didn’t fail or get blocked, it just ran slower than it should. So this isn’t a service outage or a local or network issue, which is why the hotspot, HTTP/1.1, an empty window, and turning off MCP didn’t change anything. You checked the right things.
This looks like service-side latency during peak hours. Auto routes requests dynamically, and on complex agentic tasks the planning step can take a long time under load. At 75% usage you don’t have a hard block.
What to try:
Instead of Auto, pick a specific fast model directly and compare. That helps tell if it’s Auto routing or general load.
If anything never finishes at all, not just slow but fully stuck, send the Request ID for that request and we’ll dig deeper.
Another issue that I am having, is everytime I open the conversation I was working on before everything became slow it triggers the mail app on my desktop and creates two drafts addressed to the email i was using to test my product
There are two different issues here, so let’s look at them separately.
First, about the hangs. Since single models also sometimes don’t finish complex tasks, I need the Request ID for the request that fully hung, not just slow, but never completed. I also need which model was selected and roughly what the task was. You can copy the Request ID from the chat menu in the top right corner > Copy Request ID. For the previously logged request df5d3282-c89b-4172-87ec-dd46120a2f76, it looks like it completed on our side, so it’s important to catch a truly stuck request.
Second, the Mail app and drafts issue is separate and not related to speed. It looks like that conversation includes mailto: links or email addresses, maybe in the agent output or in an open file, and they get triggered when you open the chat. To debug, can you confirm:
Does that conversation 3bf14b84-8cf0-4245-89d4-defe83e5bbe0 contain any email addresses or mailto: links?
Do the drafts get created every time you open it, or only once?
What Cursor version are you on now? In your first post it was 3.11.13, is that still the case?
To avoid mixing two bugs in one thread, please start a separate thread for the Mail app issue. That’ll make it easier to track. We’ll keep working on the hangs here.
