Note: we default stop the agent after 25 tool calls since it is an experimental feature. Please ping us at hi@cursor.com if you think we should increase this limit. If you need more tool calls, please give the agent feedback and ask it to continue.
After continuing to input, I waited for a long time but it didn’t work.
Version 0.44.9
How to break through the 25 time limit.
Personally, we are realizing that whatever al the settings are for context at $20 per month results in literally ~half of whats needed to get any sort of complex project off without several Loops of Conscious.
It would be lovely if @Cursor would try out a ‘beta’ group for some AB testing account limits – such as a group with the ‘tool call’ at ~50, ~75, ~100 – and see how much further along one can get.
I wonder if this limit could be applied to a particular workspace?
They would do well to be more transparent, but I know that AIAAAS (ai agents as a service) (as sam alltman famously said where ‘intelligence is too cheap to meter’)
Would really like to know what exactly constitutes the life of a Tool Call
It seems like running Agent mode with YOLO enabled for extended periods of time might accelerate the instability/memory issues a lot of us have been seeing lately.
I often get the agent working on a problem where it updates code, runs a curl or database command to validate the code change works, checks logs if it doesn’t, and then repeats until my stared requirements are met.
With this pattern, it can run for a long time until it resolves a problem and the only intervention is needs is a prompt to exceed the 25 tool limit. I think this loads a lot of information into the Composer - especially with repeatedly tailing the logs - and it seems to grind Cursor to a halt after awhile. Or maybe it’s all the terminal windows in Composer?
I find starting a new Composer session and/or rebooting Cursor gets it going again.
Btw, something I’ve started doing to mitigate losing context when I have to start a new Composer session is I have the model keep a log of it’s findings as it works through building a feature or debugging an issue. I ask it to write this to an MD file and continue to update as it works. I can then reference this file when I start a new Composer session due to the current session going to long or erroring out, and the new Composer session can pick up where it left off with context. This has been especially helpful when it’s doing long uninterrupted runs in agent mode which seems to exhaust the window a lot faster.
Restarting cursor and restarting the computer cannot continue 
Start a new Composer session to continue the conversation.
We are still validating what the tool call limit for the AI should be internally, but your original solution in your first post of simply sending the word ‘continue’ back to the AI should have worked fine!
The main reason for implementing the limit in the first place was to stop the AI getting stuck in a loop and infinitely using tool calls.
