This article contains affiliate links. We may earn a commission if you purchase through our links at no extra cost to you. We only recommend tools we've actually used and evaluated. Full disclosure policy here.
Cursor broke while I was mid-session on a React project last month. Autocomplete just... stopped. No error, no explanation. The little AI indicator in the status bar went grey and sat there looking useless.
I spent 40 minutes fixing it before I should have. Most of these issues have obvious solutions once you've seen them -- but when you're in the middle of a debug session and your AI coding assistant suddenly goes dark, "obvious" is not how it feels.
So here are the 10 issues that trip up Cursor users most often, and what actually fixes them.
Fix 1: Cursor Not Connecting to the AI Model
Check this first. Everything else is downstream.
Go to Cursor Settings (Ctrl+Shift+J on Windows/Linux, Cmd+Shift+J on Mac) and navigate to the AI tab. You'll see your API key configuration if you're using your own key, or your account status if you're on Cursor Pro.
If you're on Cursor Pro and using their hosted models: go to status.cursor.com and check for active incidents. Cursor has had a handful of partial outages over the past year and they're not always announced loudly. If the status page shows degraded performance on the AI inference layer, that's your answer -- wait it out.
If you're using your own OpenAI or Anthropic key: the issue is almost always the key itself. Keys expire, get revoked, or hit billing limits on the provider side. Copy a fresh key directly from your OpenAI or Anthropic dashboard, paste it in, and try again. Don't type it manually. One wrong character and it silently fails.
Also check: VPN and firewall settings. Cursor routes AI requests through specific endpoints. If you're on a corporate network with outbound traffic restrictions, or if you've got a VPN that routes everything through a country with API restrictions, that'll break the connection without an obvious error message.
Fix 2: Cursor Autocomplete Not Working
Copilot++ -- Cursor's AI autocomplete -- is the feature that makes the editor feel magical when it's on and completely forgettable when it's off.
First, confirm it's actually enabled. Go to Settings > AI and look for the Copilot++ toggle. It should be blue/on. If it's off, toggle it on and try typing in a file.
If it's already on and still not working: check what language you're in. Cursor's autocomplete works better in some languages than others. TypeScript, Python, and JavaScript have the most training data and tend to get the best completions. Niche languages or obscure config file formats sometimes don't trigger completions reliably.
Also worth checking: the file you're in. Autocomplete often doesn't fire in certain contexts -- inside comments, inside string literals in some configurations, or in very large files where the context window is saturated. Try a fresh file with a simple function and see if completions appear there.
If you're on the free tier, check your monthly completion count. You get 2,000 completions per month on the free plan. If you've burned through them (which is easier to do than you'd think), completions stop without a clear error message in the editor. Check your usage in the Cursor dashboard.
Restart Cursor if you've changed anything. Seriously, a lot of setting changes don't take effect until restart.
Fix 3: Cursor Chat Not Responding
The Chat panel (Ctrl+L) is one of Cursor's best features. When it stops responding, it's usually one of three things.
Model selection. Open the model dropdown in the chat panel. If you're on GPT-4o or Claude 3.7 Sonnet and getting slow responses or timeouts, try switching to a different model. Model availability fluctuates. Sometimes Claude is fast when GPT-4 is slow, sometimes it's the other way around. The default Cursor model (usually labeled "cursor-fast" or similar) has better rate limit headroom for Pro users.
Rate limits on the free plan. Free tier users get 50 slow premium model requests per month. Once those are gone, you're limited to faster but less capable models. If chat suddenly feels dumber or stops responding entirely, check your request count in the Cursor dashboard.
Context window overload. If you've been in a long chat session and included a lot of code context, you may have pushed the conversation toward the model's context limit. Start a new chat (the + button in the chat panel) and try again with a fresh session. Long chat histories slow down response times even before hitting the technical limit.
One more thing: if chat is responding but the responses are clearly wrong or hallucinating more than usual, you may have inadvertently selected a cheaper model or one that's running degraded. Check the model indicator and switch if needed.
Fix 4: Cursor Freezing or Crashing
Cursor is built on VS Code's Electron base, which means it inherits Electron's memory habits. Specifically: it can grow to eat significant RAM on large projects, especially with AI indexing running.
Clear the cache. On Mac: ~/Library/Application Support/Cursor/Cache. On Windows: %APPDATA%\Cursor\Cache. On Linux: ~/.config/Cursor/Cache. Delete the contents of these directories while Cursor is closed, then restart. This often fixes random freezes and slow startup.
Check memory usage. Open your system monitor while Cursor is running. If Cursor is consuming 4GB+ RAM, you're probably hitting OS-level memory pressure, especially on 8GB machines. Cursor's AI indexing keeps file embeddings in memory for fast lookups, and on large projects this adds up.
If crashes are frequent: try disabling extensions one at a time (yes, VS Code extensions work in Cursor -- yes, they can conflict -- more on this in Fix 6). The crash might not be Cursor itself.
Reinstall as a last resort. Download a fresh installer from cursor.com, uninstall the current version, and reinstall. Before doing this, export your settings (Ctrl+Shift+P > "Export Settings") so you don't lose your customizations.
Fix 5: Cursor Not Working with Specific File Types
Cursor's AI features are file-type aware. What works perfectly in a .py file sometimes doesn't trigger the same way in a .yaml, .toml, or .mdx file.
Language association check. Look at the bottom right of the editor window. It shows the detected language for the current file. If Cursor thinks your .mdx file is plain text instead of JSX, autocomplete and AI suggestions won't work correctly. Click the language indicator and manually set it to the right language. Cursor will remember this association.
AI context for config files. If you're working in infrastructure files (Terraform, Kubernetes YAML, Docker Compose), Cursor's suggestions are less confident by default. The workaround: use Chat with context. Select the relevant file content, open chat, and ask specific questions rather than relying on inline autocomplete.
Binary and generated files. Cursor will attempt to index everything in your project directory. If it's chewing through massive generated files (compiled outputs, lock files, binary assets), it slows down everything. Add a .cursorignore file to your project root (same syntax as .gitignore) and exclude the directories you don't want indexed.
Fix 6: Cursor Extension Conflicts in VS Code Mode
If you migrated to Cursor from VS Code and imported your extensions, you brought some potential conflict sources with you. Not all VS Code extensions play nicely with Cursor's AI layer.
The biggest offenders tend to be other AI-powered extensions. GitHub Copilot is the main one. If you have Copilot enabled in Cursor alongside Cursor's own Copilot++, they can fight over autocomplete suggestions and cause weird behavior -- completions firing twice, completions not firing at all, or the editor slowing down noticeably.
The rule: if you're using Cursor for AI features, disable GitHub Copilot in Cursor. They're redundant and they don't coexist gracefully.
Other extensions that sometimes cause issues: Prettier with aggressive auto-format-on-save (can interrupt AI suggestion acceptance), Tabnine (another AI autocomplete), and some Vim emulation plugins that intercept keystrokes before Cursor's suggestion handling.
To debug an extension conflict: disable all non-essential extensions, verify Cursor's AI works normally, then re-enable them one at a time until the issue reappears. That's the extension causing the problem.
Fix 7: Cursor Not Indexing Codebase Properly
Codebase indexing is what lets Cursor understand your project at a high level -- it's the difference between Cursor giving you generic suggestions and Cursor giving you suggestions that know about your specific functions, classes, and patterns.
Check indexing status. Hit Ctrl+Shift+P and type "Cursor: Index Codebase." You'll see either progress or an error. If there's an error, it's usually a permissions issue or a path that Cursor can't access.
Too many files. Cursor has limits on how many files it indexes. The practical ceiling for reliable indexing is around 50-100k files. A project with node_modules included will blow past that immediately. Your .cursorignore file should always include:
node_modules/
.git/
dist/
build/
.next/
*.min.js
New repository setup. When you open a project for the first time, indexing runs in the background. It's not instant. On a large repo, it can take 10-15 minutes before Cursor's codebase-aware features start working well. Don't judge indexing quality in the first few minutes.
Re-index manually. If the index seems stale (Cursor is suggesting things that don't exist in your current codebase, or missing things that do), force a re-index via the command palette: "Cursor: Re-index Codebase."
Fix 8: API Rate Limit Errors
This one's actually two different problems that look the same.
If you're using your own API key: You're hitting rate limits on the OpenAI or Anthropic side, not on Cursor's side. Cursor just proxies your requests. On OpenAI's free tier, you get rate-limited aggressively. A few rapid-fire chat exchanges or a big autocomplete session can exhaust your RPM (requests per minute) limit within seconds. The fix is either upgrading your API tier with the provider, or slowing down your usage. You can also switch to a different model -- GPT-3.5 and Claude Haiku have much more generous rate limits than GPT-4o and Claude 3 Sonnet.
If you're on Cursor Pro: Rate limits show as "too many requests" or slow responses during peak hours. Cursor's Pro plan gives you 500 fast premium requests per month. If you're hitting them before month end, you have three options: be more selective about when you use expensive models, use Cursor's fast model for simpler tasks, or upgrade to a Business plan.
The actual error message in Cursor for rate limiting is often vague -- "request failed" or a spinner that just stops. If you see this during heavy usage, assume rate limits before assuming a bug.
Fix 9: Cursor Not Working on a New Project or Repository
You switch to a new project, open it in Cursor, and the AI features feel dumb. Autocomplete gives generic suggestions. Chat doesn't seem to know anything about your codebase.
Almost always: the codebase index hasn't finished building yet.
But also check: did Cursor actually open the correct root folder? This matters more than it sounds. If you opened a subdirectory rather than the project root, Cursor's indexing starts from that subdirectory. It won't automatically discover files above it. Close the folder and reopen from the actual project root.
For monorepos: Cursor works best when opened at the repo root, not inside a package subdirectory. If you're in a pnpm workspace or Nx monorepo, open the root and let Cursor index the full structure. The .cursorignore file (at the root) controls what gets indexed.
Workspace-specific settings. If your previous project had specific Cursor settings configured (custom AI instructions, specific model selections), those don't carry over automatically to a new project. Check the workspace settings (.vscode/settings.json works in Cursor) if you want project-specific behavior.
For context on how Cursor compares to its competitors when everything is working correctly, our Cursor vs Copilot vs Codeium comparison breaks down which tool handles these edge cases best.
Fix 10: Cursor Slow Response Times
Slow isn't the same as broken, but it can feel that way when you're used to fast completions.
Model selection is the biggest lever. GPT-4o and Claude 3.7 Sonnet are slow during peak load. The Cursor default "fast" model is optimized for low-latency completions -- use it for autocomplete and quick tasks. Save the heavy models for complex reasoning tasks where quality matters more than speed.
Peak hours are real. Cursor (and the underlying model providers) get hammered on weekday mornings US time -- roughly 8am to noon Eastern. If you have the flexibility, running your AI-heavy development sessions in off-peak hours actually makes a measurable difference.
Your local machine matters. Cursor is an Electron app. On a machine with 8GB RAM and a spinning hard drive, it's noticeably slower than on a 16GB machine with an SSD -- not because of the AI parts, but because the editor itself runs slower. The AI requests are network calls that happen at the same speed regardless of your hardware, but everything around those requests (loading context, displaying results, running syntax highlighting) is local.
If you're on Cursor Pro and responses are still slow on simple requests: try disabling the "Always include full file context" option in Settings > AI. Cursor sometimes sends much more context than necessary for simple requests, which adds tokens and latency. Trimming context often speeds things up noticeably.
When to Contact Cursor Support
You've tried the above and something's still wrong. A few scenarios worth contacting support:
- Billing issues. Charges you don't recognize, plan upgrades not taking effect, features missing from a Pro account. These need human intervention.
- Persistent connection failures. If you've verified your API key, your network, and the status page, and Cursor still can't connect to AI features consistently, there may be an account-level issue.
- Crashes with error reports. If Cursor is crashing and generating crash reports, submit them. The dev team actually uses these.
Cursor support is at cursor.com/contact. Response times are reasonable -- usually within a business day for Pro users.
For a broader view of what Cursor does when it's working, read our Cursor AI review -- it covers the editor's actual capabilities, the Pro plan breakdown, and who it's the right tool for. And if you're still deciding whether Cursor is worth it versus staying on VS Code with Copilot, the Windsurf vs Cursor comparison is worth a read -- the two tools solve the same problem differently and one probably fits your workflow better than the other.
For Pro plan pricing and upgrades, the current plans are at cursor.com/pricing.
TechSifted is reader-supported. When you buy through links on this site, we may earn an affiliate commission at no extra cost to you. This doesn't influence our recommendations -- our editorial team evaluates tools independently.
Top comments (0)