Hey, thanks for the question.
Right now, Cursor doesn’t support directly connecting to local models via localhost. All BYOK requests go through Cursor servers to build prompts, so you need a public HTTPS endpoint.
A workaround is to use tunneling, like https://ngrok.com/ or Cloudflare Tunnel · Cloudflare One docs, to expose your local instance (Ollama, LM Studio, etc.) to the internet. Then set that URL in Cursor Settings > Models > Override OpenAI Base URL.
More setup details in these threads:
The team is aware of the request for native support for local models without tunneling. Let me know if you have any questions setting it up.