There is a lot of abuse going, people sharing account, using proxy to connect and looks like it’s one person. I saw that Cursor currently allow 10 devices per account, maybe you should consider lowering that to 2 for the Pro plan and maybe 4 on the Business plan.
As a pro user, I certainly don’t need more than 2 devices (one desktop, one laptop). I understand that some people might have more device because of their business, so the Business plan could offer that.
As people know the queue are getting long and resource are limited and I want to see the project succeed for the longterm.
That’s still far from the 10 they offer. What is happening is people are polling and sharing 1 account with 10 peoples. If they lower that to even 3 devices, it would make such practice much less attractive. Just trying to gather feedback.
How beneficial is it to share 1 account with 10 people? I’d imagine you’d hit usage limits pretty quickly, and be relegated to the same slow tier that free users get, or be force to pay for more fast requests anyway. I just don’t think this is a big concern for the cursor team, considering they’re charging based on usage anyway.
I run it on at least 3 on a daily basis. Depending on the development needs I would need to install it on another machine every now and then, but yeah, maybe 6-7 tops?
I don’t see how limiting the devices would benefit anyone. If I share my account with 10 people, I will have to start paying for extra calls every day. An idea sprung from affection, but shut down by common sense.
I don’t know why you guys all report that you need to pay, I’m over my 500 fast query, if I submit a query as we speak, there is no wait time or anything. That’s with sonnet 3.5. I never had to pay for any extra fast request. I have seen queue in the past, but not recently.
And what I’m talking about (the account sharing) is not theoretical, it’s actively being built with people actually paying for that service…
What I find puzzling is how someone can use 6 to 7 different computer in a month to do their job. Can you explain your use case, I’m really curious.
I feel Cursor is very generous, but I would like to avoid a situation where they have no choice to do like Windsurf and change the plan and eliminate the slow queries and go with a hard cap.
Well, let them. They’ll quickly surpass their limits anyway.
Oh, I can use way more. I’m building testing suites (and I do some bug hunting) for softwares that are running in Windows (practically everything since Win7 but we’re phasing that out now at least) and Linux, different dialects. Not *BSD yet, but it might be included next year.
Since we also have customer specific configurations etc I tend to do some modifications from time to time on some of the machines before the test suite has its go.
I know it sounds lazy, but we don’t test different locales unless we get a report that might be related to local number formatting errors or date representation, path formatting etc.
Here are the 64-bit Windows desktop editions that I need to test as part of our test suite:
Windows 10 Home x64
Windows 10 Pro x64
Windows 10 Enterprise x64
Windows 10 Education x64
Windows 11 Home x64
Windows 11 Pro x64
Windows 11 Enterprise x64
Windows 11 Education x64
Here are the 32-bit desktop versions:
Windows 10 Home x32
Windows 10 Pro x32
Windows 10 Education x32
Here are the Windows Server versions (no 32-bit):
Windows Server 2016 Standard x64
Windows Server 2016 Datacenter x64
Windows Server 2019 Standard x64
Windows Server 2019 Datacenter x64
Windows Server 2022 Standard x64
Windows Server 2022 Datacenter x64
Here are the 64-bit Linux versions:
Ubuntu 22.04 LTS (Jammy Jellyfish) x64
Ubuntu 23.10 (Mantic Minotaur) x64
Debian 12 (Bookworm) x64
Fedora 39 Workstation x64
Linux Mint 21.2 (Victoria) x64
openSUSE Leap 15.5 x64
Arch Linux (rolling release) x64
Here are the 32-bit Linux versions:
Debian 12 (Bookworm) x32
MX Linux x32
antiX x32
Slackware x32
Puppy Linux x32
Arch Linux 32-bit Community Edition
Ubuntu 18.04 LTS x32 (finally going end of life in April 2028, only 3½ more years now)
Linux Mint LMDE x32
And a handful of the awful Macs:
macOS Monterey (version 12) x64
macOS Ventura (version 13) x64
macOS Sonoma (version 14) x64
I do agree with you. I hope they don’t have to do that, but by then, Void has likely been launched and will be gaining speed quickly. Maybe others as well.
Im going to attempt to chain YOLOs together - whereby they open shared DB table of tools, notepads, cursorules etc from the db via .env and have a git repo where they keep all the tooling:
And this goes along with my Composer Agent Documenting, and then with SpecStory - I think Agentic Shipping will become a very nimble craft.
There is a project actively developed right now, they use proxy to share 1 account with 8 people. If the abuse keep going, at some point they will need to do like W*ndsurf and drop the unlimited slow request on the premium model. In the end, maybe that will be the best. Pay for what you use, no abuse possible.
Im in the process of building a coding game wher individuals’ agents are pointed as auth contributors to a gihub repo and they coordinate thie agents co-coding together, with different model, against the same codebase…
This isnt the same thing - but having a bunch of agents hit a codebase anonymously - but working together is an interestng thing I want to explore out of curiosity.
The only recent change is the increase in queue times for Claude 3.5 Sonnet, due to the issue we are having with getting the capacity we need from Anthropic for the LLM usage we have.