-
That is quite an old version, and there has likely been changes since to our updating infrastructure that means auto-updating will not work from that version. You can always download the latest version from cursor.com
-
No worries, there is a lot of buttons and menus in Cursor and VSCode, so easy done!
-
There is likely to be issues in your version around how this data is stored that have been fixed in newer versions. I’d highly recommend upgrading to get the best stability here.
Chat history is stored within the editor itself, and is not in a user-accessible location or format.
-
Your request about forking a composer session into a new one, but keeping the context or sections of the chat is not unique, and we have this tracked to look at on our internal feedback tracker. Hope to have more to share around this soon!
-
The best time to split of a new composer session is highly dependant on how you use it, and what you are working on, but whenever you see the quality or relevance of the AI responses start to decline, I’d recommend starting a new session then. Once you’ve worked like this for a short while, you’ll start to get a feel of when the best time to start fresh is.
The effects of a long chat are only within the chat itself, you can have 100x chats that have no effect on the AI, only the history and context from your current chat is used. However, you can always jump between different chats if that works as a good way of splitting the chats between different areas of your codebase.
Cursor does some magic behind the scenes to remove the raw token window, but there is still a point where the quality drops off. It’s not easy to “detect” programmatically, but a human can often tell.
- The file I recommend you create should be project specific, but no set format works best. Being concise but descriptive is best, high information density is best for AIs.