Model Name is not Valid "omlx"

Support Help
,
1

The free mode forces you to use AUTO so I bought a subscription (Pro). I create a custom model pointing to omlx which uses a local end point / model. I don’t want to send proprietary data to the cloud and just want to use Cursor Agent and a local model.
I thought if I bought a subscription I could use a custom model - that’s what the UI suggested when saying to upgrade to Pro.

Disappointed and would like an explanation please?

5

Hey, unfortunately I’ve got some bad news for your use case.

Pro won’t unlock local models in Agent. That’s a separate thing and it doesn’t depend on your plan. Pro gives access to frontier models like Claude, GPT, Gemini, etc that run on the providers’ infrastructure, not locally.

About custom local models in general:

  • Cursor Agent with local or custom models isn’t supported. A custom OpenAI endpoint can work, with some caveats.
  • Even if it were supported, localhost still won’t work because requests go through the Cursor backend, not directly from the IDE. You need a publicly reachable endpoint like ngrok, a public IP, or a reverse proxy. With a local omlx running on a machine without a public address, this setup just won’t work.
  • Custom model names often get rejected by validation. This is a known limitation Cursor is blocking all custom models even with a correct BYOK setup.

I get that the UI wasn’t obvious here and the “upgrade to Pro” wording could be misleading. If your main goal is not sending proprietary code to the cloud, take a look at Privacy Mode in Cursor Settings. We don’t store your code or train on it, but inference still goes through cloud providers. Fully local inference with Agent isn’t something we support right now.