How to customize user input for LLM prompt?

[lifan0127]

Is it possible to customize how user input is represented in a LLM prompt? Currently, I see that text and image files are automatically added to prompt. However, I'd like to customize this behavior to reference attachment files, structured data from message extensions etc. Thanks!

https://github.com/microsoft/teams-ai/blob/main/js/packages/teams-ai/src/prompts/UserInputMessage.ts#L55-L81

[aacebo]

no we do not currently offer an easy to use method for overriding any of the prompt sections. Technically if you wanted to the best way would be to extend the PromptManager class where they are initialized via the getPrompt method. This is an interesting idea so if you feel strongly that this should be a feature I encourage you to either submit a feature request or create a discussion/PR for this.

[lifan0127]

Thanks for your response! While I think there is a need to support more complex, potentially multi-modal input, I don't know the teams-ai library well enough to propose a concrete feature request yet.