It has been updated to 0.47.3, which is obviously improved and becomes easier to use.
Looking forward to version 0.48, update content will not have a number of claude 3.7 cooperation while editing the code base, you can pay more quick requests
Will future updates be faster and faster? By improving cursor, programming efficiency is getting faster and faster.
There’s a word on it, mistyped. When can we launch a multi-agent code at the same time?
larger codebase support!!
Call me stupid, but I like good code…that actually works…
I am still using 0.46 and have no idea what added in 0.47. But anyway, I hope to implement two revolutionary features:
Intelligent Auto Mode - Users wouldn’t need to manually specify context by @. The AI would automatically determine the current need: whether to just simpilly answer a question, modify files, add specific file context, or search the web. The current “agent” mode comes closest to my expectations, but it’s still not intelligent enough.
Advanced Voice Support - Microsoft Copilot has already added voice functionality, but it performs poorly. The reason is: that when users speak, the system fails to recognize keywords related to the workspace context, such as project related terminology, speicific project related method signatures, struct names, etc. As a CTO, I would consider abandoning the traditional “voice-to-text then to AI feedback” route, and instead adopt a smarter approach by using RAG or other technologies to directly parse user voice into structured guidance and commands before passing them to the final LLM(there could be add a bridge step, printing out the AI understood guidance in order and let user to check before sending to the final LLM). Once this feature is implemented, development efficiency would be dramatically improved.
And I suspect that the 2ed feature is more evolutionary, which has to be based on the stability of the 1st feature.
Who on earth wants to discuss 0.48 when there are still stability and output quality issues on 0.45-0.47?!
If you’re thinking about new versions, then its either a forced post, or you must be only performing the most basic of code changes on the current system, to not notice that there are fundamental issues that need resolving before something new should be considered! lol.
Honestly I do wonder sometimes.
How about make whats current reliable and consistent over a prolonged period of time.
How about full pre-release testing of updates.
How about not pandering to chasing what CoPilot is doing by being a follower, and make whats present world-class first
How about fixing complete randomness and self destructive behaviour!
How about an Accept and Reject system that actually works properly for ALL requests.
How about being able to completly reverse a responses actions in totality in EVERY circumstance! It doesnt always work and leaves traces of garbage in its destructive wake!
How about repeatability? Sometimes when the wind changes direction, mid repeatable mundane work, it decides to just completely change an established approach, and go off of complete and utter random tangents, causing the previous god knows how many spent tokens, just become a waste, overloading the system with brokenness.
How about repeatability on following AI Rules set in ALL occasions. It is still selective when it follows basic AI Rules and when it doesnt.
How about the ability to stop it losing all track of what it has been doing for the last few hours, because it went on one of its bipolar impulsive tangents, and then loses all history or knowledge of what is currently being worked on.
0.48? You must be kidding? More features to add to the cooking pot of instability? No thanks!!
On another note, Claude 3.7-Thinking is working very well this morning. But it is approaching dinner time here and it tends to go very wrong around that time. And just to clarify, well still means that it is occasionally doing all of the above. Well means it is producing quality responses in general, and not failing to connect to Anthropic or constantly reporting High Load!
Any corporation in the world would love its users to lower its expected standards. It only makes them more profitable.
I’m still on Version: 0.41.2, because it’s completely bug free. Is there a later version that is bug-free?
an MCP store maybe ? Would allow anyone to install an MCP server in “one click”, directly from Cursor
it would be great
I’d love to see better codebase indexing and/or better context management. I would be willing to dedicate whatever local resources are needed to make it work. Some sort of local Cursor-managed memory bank or knowledge graph would be amazing. The loss of context with the agent is always an issue, and trying to force it with the rules for AI or with project rules is hit or miss at best. Rules get picked up maybe 50% of the time. Even with global, those aren’t being seen in the chat if they are being used.
I also would love to see a marketplace for MCPs. There are only a few good places that aggregate them. Otherwise, you need to hunt them down with an AI search. A central place with reviews, similar to the VSCodes marketplace, would be a fantastic addition.
However, stability issues and existing bugs I’m sure are the primary concern. I am pleasantly surprised with the performance of 0.47.3 and 0.47.4. Both are working much better on macOS. I still get a few stalls here and there, but they have reduced from 10 seconds or more down to maybe 1 or 2 seconds, here and there, which honestly could be standard system interruptions at this point. I do keep 4,245 windows open.
multiple images or images at rules pls
Add support for Azure’s o1 and o3!
Cursor is on the right track now. Stability on 0.47 first.
Its already a huge improvement but needs a few more tweaks.
Also infrastructure improvements (scaling, load handling,…) and improved internal prompts for error reduction, applying code changes etc…
I think the reason that people are interested in 0.48 is because of wild ride we’ve had with 0.46 - there’s lots of anecdotes about 0.47 not really improving some of these pain points, or thats certainly my take away from the conversations on this forum - I’m not beta testing anything from Cursor for the foreseeable future - so am stuck with this 0.46 product that is literally broken.
lets guess that this is why 0.47 was created. I cant vouch for all but for me its working way better than 0.46. No major issues (for me). But yes it is different than 0.46 and people may need to adjust their prompts or rules… etc…
I think the indexing of the context base should be done via rag or um. Method that he can really search and know relevant parts of the code not only the code base but also indexed documentation
yeah thats how it works
It seems there is a lot of resonance in the comments section. Everyone is in great need of the following:
I sure hope so, I cant even run and load in a chat anymore, I guess my project is too big? I haven’t even gotten close to implementing all of the backend.
It would be nice to get word wrap on code blocks in the chat, is this planned at any point? Not the easiest experience to scroll to the right all the time
