Semantic kernel equivalent of model retry

[smokhasi]

I would like to the pass the tool call error message back to the LLM so that the LLM can correct itself. Is that possible?

Pydantic AI supports something similar using model retry

For example, I have a plugin generating sql queries which is then executed on the database and the sql query fails for some reason. I would like to be able to pass the sql error message back to the plugin responsible for generating sql queries to correct the sql query.

[smokhasi]

@moonbox3 any insight into this sort of implementation in semantic kernel?

[moonbox3]

Hi @smokhasi, admittedly we can do better at handling tool call errors in SK. We try to be as defensive as possible up until calling the function, but we're not returning the error to the model. Let me have a look at how we could cleanly support this.

[moonbox3]

I created an issue (#12946) to track this. Please follow that one for updates. Thanks.