[Enhancement]: Handle recursion limit like Cursor

[afaq-glowforge]

What features would you like to see added?

Background

Our team ran into several instances of agents crashing with a generic failure message which, after investigation logs, turned out to be because of recursion limit being reached. It took a few tries to find the upper limit we were comfortable with but still allowed important use-cases for our team to be functional (mostly agents)

Feature

When a user is making a request that hits the recursion limit, ask the user if they wish to continue via a prompt or in-conversation button. This would allow a user to acknowledge and continue.

This is how Cursor.ai works and is preferable for teams. For us it would mean we can allow our team to move faster by avoiding having to wait for an admin to update librechat and debug with them.

More details

• The configuration should be updated to opt into this behavior. Perhaps librechat.yaml can add a section under agents for this to register the soft nature of the limit. Optionally, it could also allow configuration of the step count to increment by each time

• Librechat should log instances of users requesting to surpass the recursion limit and capture the conversation ID, timestamp, and the limit they chose to exceed. Preferably all of this would be in mongodb and logs (new audit.log?)

Which components are impacted by your request?

No response

Pictures

Librechat UI,=
Agents
Librechat.yaml

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

It's a good suggestion, for now you can already set the recursion limit much higher to allow for longer runs:

https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/agents#recursionlimit