Feature Request: log *which* LLM we are talking to, not just **Assistant**

[gitcnd]

Different agents are wildly different in capabilities, and I was trying to work out which one I used with extreme success a few days ago... but no luck... the specstory didn't capture that info :-(
I don't think it's captured in the Cursor Db either :-(

It is here though! https://www.cursor.com/settings

Sorry for dropping a gnarly-hard suggestion into your issues :-)

[belucid]

No need to apologize! It's a good suggestion. Not sure that we have easy access to it, but we'll look into if we can capture which model you were working with during the AI interaction.

[gitcnd]

I tried the idea of asking the assistant itself to identify its model and variant... but all it knows is whatever Cursor told it in the prompt (which has a bug - if you pick 4o after using 3.7, cursor still sends the 3.7 prompt). the llm can guess if it's a "thinking" variant from the presence of a tag in its context, but trying to make it do that on every message is going to degrade our task context...

My use case, by the way: I switch models a lot during each chat (and, as now ones arrive, so will many others), and last week, the model I picked spat out 3200 lines of C# code into 4 files, which built and ran flawlessly first try (a windows service for full read/write access to the clipboard, running as a local named_pipe tool with an agentic self-documenting capability advertising/discovery, with test suite!). I forgot which model that was, and every attempt to replicate that awesomeness since has failed me :-(