How to get accurate LLM suggestions for fast-moving libraries?

When working with fast-moving libraries (e.g., Bevy in Rust) where the LLM wasn’t trained on the latest version, I struggle to get accurate suggestions. Even after cloning repos, the workflow feels inefficient.

What’s the best way in Cursor to make the LLM reference correct functions/APIs from specific library versions, rather than suggesting outdated ones? Any tips for a cleaner workflow?

Thanks

May I get an answer on this? (especially from the Cursor team)