Our company teams are also affected, several crashes per hour, despite the fact the crashes are very annoying, continuing always consumes extra requests. So it’s also about money. TbH, i don’t want to see anything else before this issue is fixed. Previous versions hat their flaws, but did their job without crashes.
Sorry to say so, because i like cursor, but in this state, abandoning cursor and cancelling our contracts is on the table.
Yup same problems here .. has been driving me crazy and yes i agree with comments about wasted time and wasted tokens and expense … is very frustrating .. win11 96GB RAM rtx5070ti on an external GPU host …. how can 96 GBRAM and 16GB VRAM not be enough … “DAMN YOU MICROSOFT !!!”” said in a colonel clink kind of hogans heroes voice !!! …….. To be honest I wonder if they even care about this at microsoft as they are so busy creating slop for their windows for extra fat profits …. It would be nice to know when this and if this is going to be resolved .. its a headache to the MAX .. I jam quite new to cursor and have just subscribed for the 12 month pro plan up front .. I have been really happy with things until the OOM crashes started .. I have now reduced my file overheads in the window to a minimum and still I get them .. damn annoying .. should i stay and will it be fixed ?
Hi Cursor team - your product is amazing but please fix this, we are an enterprise customer and cannot run basic generation tasks, the OOM happens after just a few mins, this was not happening prior and the team is stalling. This is urgent from my view and makes the product unusable. Thank you
I asked Cursor to patch the Otel related failure . let’s see
Here’s what the patch does across all 3 variants (src, esnext, esm):
What changed: The export() method in OTLPExporterBase now immediately calls back with { code: 0 } (which means ExportResultCode.SUCCESS) and returns, instead of forwarding to the delegate that tries to HTTP POST to Cursor’s backend. No network call is ever made.
Backups created:
OTLPExporterBase.js.bak in build/src/, build/esnext/, and build/esm/
don’t mind doing it after every update if it takes care of the OOM thingie .. the whole point is to run for hours
I sent cursor the memory leak error’s from Cursors Developer Tools → Console tab, and told them to fix the issue. this was their fixes. it was a two part prompt. I do not get anymore OTEL error’s spamming console.
Step 1: Add the missing OTEL environment variable
The most effective fix is setting the OTEL_SDK_DISABLED environment variable, which tells the OpenTelemetry SDK to completely disable itself before it even tries to export.
Environment variables (user-level, persist across restarts):
OTEL_SDK_DISABLED=true — tells the OpenTelemetry SDK to completely disable itself
OTEL_TRACES_EXPORTER=none — tells the exporter to use a no-op exporter
Additional Cursor settings added:
cursor.general.enableTelemetry: false
telemetry.enableTelemetry: false
telemetry.enableCrashReporter: false
**
Layer 1 — Environment variables** (throttle the BatchSpanProcessor):
Variable
Value
Effect
OTEL_BSP_SCHEDULE_DELAY
2147483647
~25 days between export attempts
OTEL_BSP_MAX_QUEUE_SIZE
1
Queue holds only 1 span
OTEL_BSP_MAX_EXPORT_BATCH_SIZE
1
Batch size of 1
OTEL_BSP_EXPORT_TIMEOUT
1
1ms timeout (instant failure)
OTEL_METRICS_EXPORTER
none
No metrics exports
OTEL_LOGS_EXPORTER
none
No log exports
Layer 2 — Patched OTLPExporterBase.export() across all 3 builds (CJS, ESM, esnext):
Now immediately calls resultCallback({ code: 0 }) (success) instead of delegating to the actual exporter
No data is serialized, no HTTP connection is created
Layer 3 — Patched sendWithHttp() across all 3 builds:
Now immediately calls onDone({ status: 'success' }) instead of making any HTTP request
Belt-and-suspenders in case anything bypasses Layer 2
Required: Full restart of Cursor
Close all Cursor windows completely (check Task Manager to make sure Cursor.exe is fully dead), then reopen. The patched files are loaded at startup.
Caveat
These patches to Cursor’s node_modules will be overwritten when Cursor auto-updates. If the errors return after an update, the patches will need to be reapplied. You already have "update.mode": "none" set, which should prevent auto-updates, so you should be fine for now.
With the latest Cursor update, my issues are now fully resolved. I no longer experience crashes, and the system rarely gets stuck anymore. I have to give real credit here — I was already very close to switching to another provider.
In addition, I’ve slightly adjusted my workflow:
I export my chat once per week and store it locally in my filesystem
When starting a new agent, I explicitly instruct it to review the previous chat export and continue development based on that context
I delete old agents instead of archiving them, to keep things clean and avoid buildup
Since doing this, everything has been running much more smoothly.
So honestly, great improvement — well done to the team.
For anyone still experiencing issues:
I recommend backing up your chats and doing a complete clean reinstall of Cursor. If the problem still persists after that, then it’s likely a different issue than the one I had.
From my experience now, the latest version can be used reliably again.