> Would there be a means of limiting which services and functions each query can call manually to impose more specified guardrails?
By default, all functions which are registered are available for use in a Run. It is also possible to explicitly attach a subset of functions to any given "agent".
> how is the “human in the loop” thing enforced?
It is possible to return what we call an Interrupt from within your function code [1]. When this is received, the run is paused until it has been satisfied. Inferable's control plane handles this not the model in case it isn't clear.
> Would there be a means of limiting which services and functions each query can call manually to impose more specified guardrails?
By default, all functions which are registered are available for use in a Run. It is also possible to explicitly attach a subset of functions to any given "agent".
> how is the “human in the loop” thing enforced?
It is possible to return what we call an Interrupt from within your function code [1]. When this is received, the run is paused until it has been satisfied. Inferable's control plane handles this not the model in case it isn't clear.
[1] https://docs.inferable.ai/pages/human-in-the-loop