I kind of realized yesterday that the agent chat component is somehow designed to re-trigger the linked query until it gets the results.
I guess another way of approaching the challenge of passing context from apps to agents would be to enable the option to link 2 queries, one for invoke and for getLogs. As such, we could then manipulate the invoke query and pass in additional context from the apps.
I'm aware Kent already suggested this other solution, but just wanted to share my thoughts in case it helps.
Interesting idea, it would make sense to better break up the logic of 'what' the LLM behind the agent should be trying to do and 'what' the agent has to work with to unterstand context and make decisions.
How are you imagining the breakdown of inputs/outputs for the 'invoke' query and the 'getLogs' query to look?
Are you thinking we need additional support for getting logs from the agent's own run to be able to give the agent contextual understanding of why a query or tool it is invoking may not be working?