I’m still getting the following error message: “Composer relies on custom models that require a Pro or Business subscription. Please disable API keys or use a Pro or Business subscription.”
I have been using the chat feature with just my API keys up until this update
ZScaler always gives these types of issues. Normally I can just register the cert chain in the keystore for the system having issue but does not seem like Cursor has that option
@danperks There is one major ux issue in the Chat when on Edit or Agent mode (not on Ask mode for some reason). When I try to “deleteRight” character using ctrl+d the whole Chat panel closes completely on macOS! Also happens with cmd+d. There is no keybinding visible in “Preferences: Open Keyboard Shortcuts” which I could unbind to fix this! Please remove those hard-coded keybindings since using ctrl+d is very common in unix systems.
I have been wanting to mention this since quite some time and this “agentic” mode first is again creating confusions - “what is going to be the pricing implication” - shall we not have chats free ?
<For someone like cursor rolling out real nice product updates - and being a new developer based in India where the dollar prices really hurts, cursor pricing mentions come across as “in-transparent” at the least, or may be I am not able to find out right resource !! someone plz help me understand this in simple formats…>
Current UI bottom most elevated button of “chat” servers no purpose apparently. So it is further confusing.
So please temme/confirm the below -
everything is a chat so nothing like composer. We can run agents on any model. ?
Agent can execute task, for code snippets from chat - I have to do it manually ?
There are mentioned models as free 4o mini, Cursor small, and Deepseek V3 - will it be charged for agentic task if engaged ?
Does cursor also charge out of premium requests for tab completion ?
I seem to be having a lot and lot of premium requests leakage - changing 4 mid size markdowns (~300 lines at max) and two rounds of iterations on sonnet 1022 consumed ~60+ premium tokens. Is this normal ?
The premium tokens consumed quota updates real time for Sonnet and GPT. It doesnt do so for Gemini or others ? Is that so ? Because I have often realized lost tokens the next day ?
In general - what exactly constitutes one premium token ? what exactly constitutes a large request ? // Earlier, sonnet was able to create ~5-12 smallish files for me in 1 token. Now it stubbornly stops after 2-3 and the token consumption is always like through the roof…
7B: If i keep asking it to refer codebase after (which has been indexed) - does it constitute a large request ?
7C: Do files referenced in a message and not-edited/slightly-edited/majorly-edited in next message reference count towards “large requests”
**
Plz guide me on these. Of late, it becomes a splurge everytime I touch any agentic mode.
** Also, do put some PPP consideration on pricing. A $ pricing in USA is ~12 times more burden than for an Indian developer.
We’ve kind of worked to a hybrid version here, which is neither chat or composer in its original form! All the modes should be much similar to each other, to summarise:
Agent - Fully featured, with tools
Edit - No tools, can only read and write to files
Ask - No tools, no read/write - discussion only
Yes, if you use the ‘Ask’ mode, Cursor will not execute any edits or tools. We primarily use Agent here unless you explicitly don’t want it to make any edits - we find the agent is smart enough to know if you just want to discuss, or if you actually want to make edits and changes.
You aren’t able to use any of the free models in Agent mode. If you try, you will be prompted to switch models or switch to Edit mode, which doesn’t have any tools.
Nope, Tab is free and unlimited with any paid plan.
This doesn’t sound right. Each prompt you do with a premium model uses 1 usage from your default allowance of 500. Free models and Tab do not contribute. If you find the AI is stuck in a loop, or not performing well, it may be worth starting a new chat session to start fresh!
The Gemini models are actually free, so shouldn’t contribute to your allowance
A premium request is exactly that - one query to a premium model. The only limitation here is the 25 tool call limit. If, in it’s reply, the AI uses 25 tool calls, it will be stopped to avoid it getting in a loop. If you choose to continue it, this will also use another premium request. Besides that, there is nothing that should cause >1 premium request to be used from your allowance.
7B. Again, if you ask anything to the AI, it will use 1 request, regardless of if it’s a new query, a follow-up question, or even responding to a question the AI has asked you.
7C. The amount of files you @ into the model, and any edits it chooses to do does not constitute any more usage. The only downside is that if you @ too many files, it may loose focus on the details within them - this is a downside of the underlying models though, not Cursor itself!