I am currently using Nvidia NIM APIs in Cursor. A few days ago it stops generating response, only showing “planning next moves” for seconds before vanishing (I addressed this because I have browsed some posts in the forum, yet nobody has similar problems). I have overrided the base OpenAI Base URL to https://integrate.api.nvidia.com/v1
For more information, I am living in China and connecting to a remote Linux server via SSH. I have deployed VPN connections in the server (curl command works well no matter the VPN is used or not). All official Cursor models are operational (both in Auto mode and specifically selected mode).
And I have undergone some Cursor updates in the last few days, I am not sure if it is related.
Hey, I see you’re using z-ai/glm4.7 via Nvidia NIM. Since curl works but Cursor doesn’t, this is a known issue with BYOK and custom endpoints.
Can you try this:
Get error details via Developer Tools:
Help > Toggle Developer Tools
Console tab
Send a request in chat (so you see “Planning next moves”)
Check if there are any error messages in the console
A screenshot or the error text will help us escalate
Check the Network tab:
Developer Tools > Network
Filter: fetch/XHR
Send a request and see what response comes back from integrate.api.nvidia.com
It might be 200 OK but Cursor isn’t parsing the response correctly
Which Cursor version worked?
You said it stopped working a few days ago
What was the last version where it worked?
This helps narrow down what changed
Context: The team is already tracking a similar issue with Nvidia BYOK endpoints. Most likely Cursor isn’t handling streaming responses correctly, or it doesn’t match the response format from the Nvidia NIM API.
If you can grab the error logs from Developer Tools, send them here and it’ll help the team fix it faster.
A strange bug emerged when testing the API. Days ago, the Cursor IDE sometimes turn off the API Key, which was reported by other users. When submitting request with API not enabled, a error window would show indicating the current model is not supported by current plan. However, I have noticed that along with this incident, even with the API not enabled, the error window will never show up, and the “planning next moves” vanished, just like when the API is enabled.
Please see vscode-app-1768591134600.txt (10.3 KB). The original .log file from the developer tool was not accepted thus I changed it to .txt.
Please see vscode-app_redacted.txt (49.0 KB). The original .har file from the developer tool was neither accepted thus I changed it to .txt. Along with <1>, I have tested two request, one with API enabled and one without. In the .har file, the first FOUR entries were logged seconds after the first request was submitted, and the last three was after the second one. Nothing was from integrate.api.nvidia.com.
According to Cursor change log, the latest updates are Jan 8 (CLI) and Dec 22 (2.3). It doesn’t work with the current version 2.4.0 (both in default and nightly), but it did work in the previous versions, although from time to time. More specifically, I can submit a few number of requests via custom API per day, and get some errors about provider (400 or 404) occasionally.
