life update: i joined Anysphere to work on Cursor pls send me feature reqs + feedback

@anysphere @cursor_ai Welcome Nick!

@anysphere @cursor_ai 🙏🙏🙏

@anysphere @cursor_ai lfg great to have you

@anysphere @cursor_ai 🙏🙏🙏

@anysphere @cursor_ai Support local models without going through your proxy server. We require 100% control and custody of our data.

@anysphere @cursor_ai let me speak to cursor as if I was speaking to a PM (using voice) ... once this is done the world will change ...

@thetoolists @anysphere @cursor_ai i like this idea a lot

@anysphere @cursor_ai me too ...

@anysphere @cursor_ai feature request: - show which model was in my luck when using auto

@prarthanagarwal @anysphere @cursor_ai ooo

@anysphere @cursor_ai There is a bug in cmd+k, when you give some json dict in input and expect it to modify output some dict in output with other code it breaks in between

I have a simple request. I abandoned cursor because I couldn't do line-by-line Autocompletion. (It's broken in cursor) Only non-serious engineers accept TAB completion. I just want to understand every line as I auto complete them, because I'm responsible for that line in production

@anysphere @cursor_ai really? what's stopping you from reviewing each line before you hit TAB for the whole thing? isn't it easier when you can see the entire preview, instead of one line at a time?

@anysphere @cursor_ai Because I don't want to accept 100% of TAB suggestions and there is no way to tell Cursor I like line 1 and 3 but 2 ,5, 6 are wrong. So, I'm constantly deleting things that are wrong, which is bad for my flow

@anysphere @cursor_ai ahh i see

@anysphere @cursor_ai thanks! see my other reply. It's an extremely simple bug fix on your end Make this key-binding work editor.action.inlineSuggest.acceptNextLine

@anysphere @cursor_ai Multiple models per request (a la lmarena) so we can pick between few options and continue from there. Also finetuning using thumbs up/down on the entire team

@anysphere @cursor_ai would be interested in general best of N too? and have you tried using thumbs up/down within the existing agent sidebar?

@anysphere @cursor_ai Thumbs do something now?

@anysphere @cursor_ai We need the ability to see how much code was written by AI per day.

@anysphere @cursor_ai have you taken a look at ? should have a bunch of data

@anysphere @cursor_ai 🫡

@anysphere @cursor_ai That’s awesome!!!

@anysphere @cursor_ai CONGRATULATIONS!!!!

@anysphere @cursor_ai 🙏🙏🙏

@anysphere @cursor_ai Congrats!!

@anysphere @cursor_ai thanks!!!

@anysphere @cursor_ai More explicit task manager view like how v0 does it for complex tasks. I pretty much live between the two and a bit more similarity will enhance my productivity even more :)

@anysphere @cursor_ai I registered with my student email (.edu) which meets the conditions (Hungarian university) but only got Free Pro Trial which ended after 2 weeks. I reported the issue to Cursor support, no reply. Couldn't get student discount at all

@anysphere @cursor_ai Here’s one big fix for your first impact: restore to checkpoint must use the editor’s undo/redo stack. Currently after restoring to a checkpoint, you lose that stack!

@anysphere @cursor_ai That’s sick. Congrats!

@anysphere @cursor_ai thanks!!

@anysphere @cursor_ai congrats nick!

@anysphere @cursor_ai thanks!!

@anysphere @cursor_ai

@anysphere @cursor_ai Congrats! 🎉

@anysphere @cursor_ai make it use less ram tyvm

@anysphere @cursor_ai congrats! would there any support for the new (cheap) Chinese models (Qwen, new Kimi, Deepseek) with the Agent? They are decent, affordable models.

@anysphere @cursor_ai will look into it!

@anysphere @cursor_ai Hey mate just one word 👏

@anysphere @cursor_ai how's this animation made? it's sick

@anysphere @cursor_ai 🔥🔥🔥

@anysphere @cursor_ai congratulations

@anysphere @cursor_ai Congrats, Nick!

@anysphere @cursor_ai Legend

@anysphere @cursor_ai Three modes: no todo lists, sometimes todo lists, and always todolists. Cause I love those todo lists it makes

@anysphere @cursor_ai what model was used when agent run with auto

@anysphere @cursor_ai yea that would be nice actually

@anysphere @cursor_ai The ability for AI models to read PDFs. This is very important for certain programmers. Needs to be able to properly parse 1000+ page PDFs.

@anysphere @cursor_ai can you say more about this use case? and what UX you would want?

Sure. I work with microcontrollers. Right now this one: There are lot's of specifics when it comes to all features and peripherals. How do I setup the clock system, how do I set up the RTC, timers, UART, etc... Current models guess a lot, and it's almost hopeless to get those specific things correct if you don't feed it the datasheet. So for those, I right now use straight up Grok or Chat GPT in their web interfaces for those things. Or perhaps Gemini CLI that can also read PDF's, but only max 1000 pages. I don't think UX could help much, if it's not about selecting parts of PDF's I want it to consider I guess.

@anysphere @cursor_ai so perhaps drag and drop the PDF into the chat? or an MCP server connected to the PDFs that the agent could ask questions?

@anysphere @cursor_ai Yeah, something like that. I have not delved into MCP's yet. But I guess that could work. 👍

@anysphere @cursor_ai Congrats 🎉

@anysphere @cursor_ai huge

@anysphere @cursor_ai Congrats!