Persistent Connection Error

[randalljam (Randy True)]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

SEV: Cursor AI unusable — all models fail (incl. Composer1) with error 57 “Provider Error” + [resource_exhausted] — Request ID inside

I’m seeing a blocking AI outage in Cursor: every model fails, including Cursor’s internal model (Composer1). This stopped mid-thread, persisted for ~2+ hours With dozens of retries on multiple threads, on multiple repos, And it’s still persisting the next morning, 12 hours later. Environment

Region: Portola Valley, California, USA

VPN/Proxy/Firewall: None

OS: macOS Sequoia 15.3.1

Cursor version: 2.4.21

Updates: “Check for Updates” shows no updates available (already on latest for my track)

Symptoms

New chat threads also fail immediately (“test is this working” → same error)

Models tried (all fail): GPT 5.2 Codex, Composer1 (Cursor internal), plus other providers/models that previously worked for weeks

Error is repeatable on every retry

Request / error details (verbatim)

Request ID: c04c3199-da9b-48dd-9870-530794b72f83

{“error”:57,“details”:{“title”:“Provider Error”,“detail”:“We’re having trouble connecting to the model provider. This might be temporary - please try again in a moment.”,“isRetryable”:true,“additionalInfo”:{},“buttons”:,“planChoices”:},“isExpected”:true}

[resource_exhausted] Error

(Stack trace available; points into workbench.desktop.main.js)

What I’ve tried (no improvement)

Restarted Cursor

Command Palette → “Developer: Reload Window”

Set HTTP Compatibility Mode → HTTP/1.1 (disabled HTTP/2) and restarted

Ran Cursor Settings → Network → Run Diagnostics → all checks passed / green (I can attach a screenshot)

Attachments

Screenshot of Network → Run Diagnostics results (all green)

If useful, I can also attach logs from ~/Library/Application Support/Cursor/logs

Given that this impacts all models including Composer1, can you confirm whether there is an ongoing backend/provider issue and share any ETA or mitigation steps?

Steps to Reproduce

It’s persistent and still happening.

Expected Behavior

The basic functionality of Cursor - getting an AI response.

Operating System

MacOS

Version Information

Version: 2.4.21
VSCode Version: 1.105.1
Commit: dc8361355d709f306d5159635a677a571b277bc0
Date: 2026-01-22T16:57:59.675Z
Build Type: Stable
Release Track: Default
Electron: 39.2.7
Chromium: 142.0.7444.235
Node.js: 22.21.1
V8: 14.2.231.21-electron.0
OS: Darwin arm64 24.3.0

For AI issues: which model did you use?

Opus4.5
GPT5.2-Codex
Composer1

For AI issues: add Request ID with privacy disabled

Request ID: c04c3199-da9b-48dd-9870-530794b72f83
[resource_exhausted] Error
LTe: [resource_exhausted] Error
at kmf (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9095:38337)
at Cmf (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9095:37240)
at $mf (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9096:4395)
at ova.run (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9096:8170)
at async qyt.resume (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:34190:60450)
at async Wpc.streamFromAgentBackend (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:34239:7695)
at async Wpc.getAgentStreamResponse (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:34239:8436)
at async FTe.submitChatMaybeAbortCurrent (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9170:14575)

Additional Information

See above

Does this stop you from using Cursor

Yes - Cursor is unusable

[deanrie]

Hey, thanks for the report.

I’ve seen similar cases with error 57 + resource_exhausted. It’s often caused by third-party API key settings or a Base URL override.

Could you check:

1. Cursor Settings → Models

   • Make sure Override OpenAI Base URL is turned off
   • Check if any third-party API keys were added (OpenAI, Anthropic, etc.). If yes, try removing them temporarily

2. If you do have third-party keys, try fully signing out and back in (Cursor Settings → Sign Out, then Sign In)

3. Create a new test project/workspace and try sending a simple message in chat there

In one similar case, the issue was that the user accidentally enabled a third-party GLM-4.7 key. After disabling it, everything worked again: Cursor not working! - #9 by user1717

Let me know how it goes.

[randalljam (Randy True)]

Thank you for the fast response. It’s working again! I’m back in business, which is awesome.

So I logged out and logged back in and then also in the settings turned off the OpenAI API key. I didn’t blow out those API keys. I have one in for OpenAI and Anthropic. I did not have the override OpenAI base URL turned on, so I left it turned off.

I guess the key was logging out and logging back in. I’m embarrassed that I didn’t think to do that. I mean, it’s right up there with, you know, seeing if it’s plugged in in troubleshooting. And strange that ChatGPT didn’t tell me to do that either.

Why doesn’t Cursor have an AI help that can give me this answer? Instead of stuck going to chatgpt, cutting and pasting error codes, and getting unhelpful responses. Maybe y’all are working on an AI help tool and the progress has been truly amazing overall at Cursor, but -If y’all can turn agents loose for a week straight to make a 3 million line browser, then you should be able to get a button and cursor that just says, “Help me fix my error.” And it draws from the latest forum posts and cumulative knowledge of the expert support team, which you clearly just showed Cursor has . Thanks again, and score one for humans!

If there’s anything like this and I’m just missing it, then please let me know. I guess maybe y’all’s view is that this bug report in the forum is it. I also sent an email to [email protected]. But I don’t think I’m unusual and I would much rather just do this with an integrated AI help tool in the Cursor app - That way you don’t have to cut and paste any error codes or settings or my cursor version. It should just be able to automatically read all that.

Okay, enough rant and feedback - back to work on my project. Thanks again and keep up the great work!!

-Randy