Sometimes, during a longer conversation, the agent forgets which technologies I use. For example, I work with Tortoise ORM, but it suddenly starts suggesting SQLAlchemy instead.
I would like an option to pin “facts” or “constants” in the conversation to ensure better customization.
Each constant should have an enable/disable checkbox, allowing temporary deactivation without the need to remove it permanently.
Facts should be pinned above the conversation.
What you think?
1 Like
This would be awesome, in the meantime using notepad can be useful
The user should be able to pin any prompt.
In this image I’ve added a little pin icon to make the message sticky. Text scrolls under it to clarify that it’s pinned and will be sent with any message.
Certain AI systems support “facts” or “memories”. However, since Cursor allows us to use different AI backends, this would be cumbersome to maintain.
Most AI systems require you to resubmit the entire context, and the have a size limit, so agents use a scrolling window. By allowing the user to select which messages they want pinned to the “top” of the block sent to AI, they can keep certain goals in context, preventing Cursor for running off on a tangent.
The user should be able to unpin them as well. Sometimes you want to go on a tangent. Sometimes you’ve finished your original task, or you like the direction something is going. Unpinning those message items allows you to move in the new direction.