[Enhancement]: Remove <think> part from title generation

[ErykCh]

What features would you like to see added?

If I have localy only thinking model run and this model generates me also titles, than I have something like this:

there should be option to filter out this part.

Also new models has ability to switch off thinking as for example Qwen3 with /no_think, so there should be easy way using configuration file librechat.yaml to set own prompt for title generation or be able to add additional information to the prompt.

More details

Which components are impacted by your request?

No response

Pictures

No response

Code of Conduct

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

[danny-avila]

For now, I would not even recommend using any reasoning models for title generation. That title is likely also cut off because of a max_tokens limit, so there's filtering the think tag would just make it empty.

Customizing title generation will be implemented but you can already customize the model used:

https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/custom_endpoint#titlemodel

[SurealCereal]

Using a non-thinking model is definitely the best option at present (I chose qwen2.5:0.5b which is more than good enough for creating titles & summaries). I think a small or specialized model makes more sense, although it would be useful to use the default model with reasoning mode disabled for low VRAM situations so that model-swapping is not needed. Thanks for the awesome chat UI!

[petervcook]

Agree, would love to prefix 'no_think' on the summarization prompt