Usage limit on free version?

Support Help
1

I’m running Cursor on Win 11 along with Ollama and phi. I got a message I’d reached the limit. I thought in this configuration, there were no limits. Please explain.

2

hi @Sam_Stamport Cursor Free account still uses Cursor servers to communicate with AI even if you expose your local Ollama setup publicly.

Could you share more how you use Cursor with Ollama?

  • Is the API URL configured correctly?
  • Does the model dropdown in Chat show phi model?
  • Which mode have you been using? Agent or Ask?
3

Thank you for replying promptly.

As far as I know, the API is configured correctly.

The AI model shows Auto.

When does my limit refresh?

Why does Cursor need AI in the cloud if I’m using Ollama & phi?

Chatgpt suggested this as a workaround. I’m in the process of implementing it.

Cursor by default will call its cloud models unless told otherwise.
Right now, there’s no single “offline mode” toggle, but you can disable cloud entirely by:

  1. Settings → Features → AI Provider — choose “Custom API” and point to Ollama:

    yaml

    CopyEdit

    Base URL: http://localhost:11434 API Key: (leave blank) Model: phi3

    This prevents requests from going to OpenAI or Anthropic.

  2. Disable telemetry:

    • Settings → Search “telemetry” → Turn off all logging & usage reporting.

  3. Block cloud at network level (paranoid mode):

    • Use a firewall rule (Windows Firewall, Little Snitch on Mac) to block *.cursor.so except localhost.

    • This ensures even if a plugin tries cloud fallback, it can’t send data out.

:white_check_mark: Bottom line:

  • If you fully switch to VS Code + Continue + Ollama, you get Cursor-like AI features with no usage caps.

  • If you stick with Cursor but want to be offline/local only, you must change provider to localhost and optionally block network calls.

If you’d like, I can give you a step-by-step “Cursor local-only” recipe that keeps its UI but forces it to behave exactly like a local-only VS Code + Ollama setup. That might save you from switching entirely.

Ask ChatGPT

Cursor by default will call its cloud models unless told otherwise.
Right now, there’s no single “offline mode” toggle, but you can disable cloud entirely by:

  1. Settings → Features → AI Provider — choose “Custom API” and point to Ollama:

    yaml

    CopyEdit

    Base URL: http://localhost:11434 API Key: (leave blank) Model: phi3

    This prevents requests from going to OpenAI or Anthropic.

  2. Disable telemetry:

    • Settings → Search “telemetry” → Turn off all logging & usage reporting.

  3. Block cloud at network level (paranoid mode):

    • Use a firewall rule (Windows Firewall, Little Snitch on Mac) to block *.cursor.so except localhost.

    • This ensures even if a plugin tries cloud fallback, it can’t send data out.

:white_check_mark: Bottom line:

  • If you fully switch to VS Code + Continue + Ollama, you get Cursor-like AI features with no usage caps.

  • If you stick with Cursor but want to be offline/local only, you must change provider to localhost and optionally block network calls.

If you’d like, I can give you a step-by-step “Cursor local-only” recipe that keeps its UI but forces it to behave exactly like a local-only VS Code + Ollama setup. That might save you from switching entirely.

Ask ChatGPT

Cursor by default will call its cloud models unless told otherwise.
Right now, there’s no single “offline mode” toggle, but you can disable cloud entirely by:

  1. Settings → Features → AI Provider — choose “Custom API” and point to Ollama:

    yaml

    CopyEdit

    Base URL: http://localhost:11434 API Key: (leave blank) Model: phi3

    This prevents requests from going to OpenAI or Anthropic.

  2. Disable telemetry:

    • Settings → Search “telemetry” → Turn off all logging & usage reporting.

  3. Block cloud at network level (paranoid mode):

    • Use a firewall rule (Windows Firewall, Little Snitch on Mac) to block *.cursor.so except localhost.

    • This ensures even if a plugin tries cloud fallback, it can’t send data out.

:white_check_mark: Bottom line:

  • If you fully switch to VS Code + Continue + Ollama, you get Cursor-like AI features with no usage caps.

  • If you stick with Cursor but want to be offline/local only, you must change provider to localhost and optionally block network calls.

If you’d like, I can give you a step-by-step “Cursor local-only” recipe that keeps its UI but forces it to behave exactly like a local-only VS Code + Ollama setup. That might save you from switching entirely.

Ask ChatGPT

4

@Sam_Stamport sorry but that is not what I asked about. Your question was about the limit and I assume you may have used our regular AI models instead of phi from Ollama.

Could you share your log from Dashboard Usage > All Raw Events as that would show which model you used.

http://localhost:11434 is not a supported API for Cursor. We can only access APIs that are publicly reachable.

Could you please share a screenshot of how you use Cursor (Agent model selection)

6

I am working on this handbook which should answer all your questions.

AI_IDEs_Real-Life_Handbook_Sam_Stamport.docx (31.7 KB)

7

Thank you but I can not open attachments here. Feel free to post image here and provide succinct answers.

8

Bottom line here is after trying VS Code, Windsurf, VS Codium, and Cursor I’ve decided to go back to VS Code. Of the four, VS Code is the best documented therefore I get more help from chatgpt. I know Microsoft is collecting data. Windsurf comes in second, but I won’t use it because it’s future is uncertain.