Cursor on Intellij has multiple options to select different LLMs to use But if we apply a rule to use spesific LMM for spesific task type with mdc file at rules directory, when using cursor on intellij we are not sure which llm is in effect , except for the manually selected LLM information. So its a little bit ambigious that User are not aware which LLM is in use while using it when a rules apply to use ceratin LLMs.
Hi @uysalm Thanks for the post. Project rules (.mdc files under .cursor/rules) are instructions the agent reads. They do not switch which model is running. If a rule says “use model X for this kind of task,” that is prompt guidance, not a model change. The model in the selector is the one in effect.
Asking the agent which model it is also is not a reliable check. It can follow the rule text instead of reporting the selector. In your screenshot, Composer 2.5 in the picker is the source of truth for that chat.
On IntelliJ, choose the model you want in the Cursor agent model selector before you send the request. If you want Cursor to pick a model by task type, use Auto. Auto can change from turn to turn, and the specific routed model is typically not exposed to the user.
Thanks for the Reply Kevin. I understand Auto already chooses/balances which model is used. That helps a lot, and we do use it.
The gap for us is that Auto’s routing doesn’t follow our team rules (which model for which task type) or we are not aware of it, and we usually don’t see which model was actually used for a given turn. Tasks also change mid-chat, so even Auto can pick something that doesn’t match the work at that moment.
If Auto could eventually take .mdc routing hints into account, and show the model used per response, that would close the loop for us. Manual selection could stay as an override. My main purpose is to avoid manual LLM selection by user since it might happen that user can forget to change the LLM and use a simple LLM for a complex task .Its prone to mistakes.
Would that be something worth considering?
