Hello, forum members! 
I’m currently working on an educational project in Java backend, using technologies like Spring, Hibernate, JDBC, and MVC.
My idea is to take a ready-made task template and gradually refine it with each new task, building functionality according to the conditions. For this, I’m actively using Cursor AI — a tool that, in theory, should simplify working with code. I open the project in Cursor AI, run the Ctrl + L command with the @codebase annotation so that the AI can analyze all my code, and initially, everything goes well. However, I’ve encountered several problems that prevent me from fully utilizing this tool, and I would like to discuss them with you.
Context Problem
The first thing that stands out: if I work on the project for a long time in one session, the AI starts to lose track — it forgets the beginning of our interaction and the details of the context. This creates inconvenience because I have to re-explain what I’m doing and why. But it gets even worse when I save the project, close Cursor AI, and then open it again. In such cases, the AI completely loses all history and doesn’t remember anything we discussed earlier. As a result, every time I start from scratch, which significantly slows down progress.
Question About Models
Another point that confuses me: in Cursor AI, I only see junior language models like o1-mini, Claude 3.5, o2, and others, but there are no senior versions (for example, o1 or o3). I suspect that senior models might handle the task of context retention better, but I don’t have access to them. Is this a limitation of the free version, or are senior models also unavailable in the paid subscription?
Paid Subscription and Limits
Speaking of subscription: I have a paid version of Cursor AI, and recently I received a notification that I have used 250 out of 500 fast queries. This puzzled me a bit. What will happen when I reach the limit of 500 queries? Will the work just stop, and I’ll have to buy additional queries? Or will the subscription switch to some limited mode, for example, with reduced speed? I don’t fully understand how the limit system works within the monthly subscription, and I would like to clarify this point.
Additional Features and Versions
I’ve heard that Cursor AI has features like Chat or Notepads. Can they somehow help preserve context between sessions or at least within one session? If so, how do I use them correctly? I also came across discussions here on the forum where some users suggest rolling back to previous versions of Cursor AI. I currently have version 0.46.8. Could switching to an earlier version solve the context retention problem? If anyone has tried this, please share your experience — does this approach work, and which version should I try?
My Questions
To make everything as clear as possible, here is a list of what I would like to know:
1. Are there ways to configure Cursor AI so that it better remembers the context within one session? Maybe I’m missing some settings or commands?
2. How can I make the context persist between sessions when I close and reopen the project?
3. Can senior models (for example, o1 or o3) improve the context retention situation? And if so, how can I gain access to them in Cursor AI?
4. Is it worth relying on features like Chat or Notepads to solve this problem? How do I apply them in my case?
5. What will happen when I use up all 500 fast queries in the paid subscription? Will I need to buy more, or does the limit reset somehow? How does the monthly subscription work in this regard?
6. Will rolling back to a previous version of Cursor AI (I currently have 0.46.8) help? If so, which version would you recommend?
I would be very grateful for any ideas, experiences, or clarifications. I’m especially interested in hearing from those who have faced similar difficulties while working with Cursor AI and found ways to overcome them. Thank you in advance for your help! 
