Hey all,
I’ve been using Cursor for a while now and recently noticed some concerning behavior that I feel deserves a discussion:
1. Cursor becomes unusable during long chats
When I have a long conversation or a thread with lots of context, Cursor starts to lag badly. CPU and memory usage spike, the UI freezes or becomes glitchy, and sometimes it just force-closes. But the moment I close the thread and start a new one, it works fine again.
Feels like it’s intentionally killing longer sessions. Could be memory issues, but it’s suspicious how repeatable and “fixed” it becomes once the context resets.
2. Auto Mode is suddenly using paid tokens
Auto used to feel like it was using local or at least free quota — now it’s either:
- Pulling from my paid requests,
- Or bypassing local models completely.
Even simple tasks now seem to hit GPT-4 and show up on my usage history — even if I didn’t explicitly choose it. And if I’m right, that’s a silent switch from how Auto mode worked before.
3. Local refills missing for days
I haven’t received a local quota refill in 3+ days. Not even after inactivity or light usage. Previously, it was fairly consistent. I suspect either:
- Local refills are being slowed or quietly turned off,
- Or Auto is just defaulting to paid usage to burn tokens.
The Big Picture
Put this together and it feels like Cursor is intentionally nudging (or forcing) users to consume more paid requests while making it hard to verify what’s local vs cloud.
This wouldn’t be such a big deal if they were transparent about it, but I haven’t seen any clear updates or changelogs explaining this shift.
Is anyone else experiencing:
- Cursor lag or freezing on long threads?
- Auto mode consuming paid usage when it used to be free?
- Local usage/refill not working?
- Overall degraded performance in Auto replies?
Would love to hear from others — maybe I’m wrong, but something feels off.