Retry

The retry function re-runs the current Agent. This is useful for when the Agent initially generates a bad response. Due to the undeterministic nature of LLMs retrying a request can often yield a better response.

Example "Retry when quality check fails":

agent {
    name = "weather"
    filterOutput {
       if(!qualityCheck()) {
           retry(max = 3)
       }
    }
}

When the retry function is called, the agent will be executed with the same parameters and state as the original run, with the only difference being that an instance of RetrySignal will be added to the context.

The RetrySignal bean can be accessed with the get function and will contain the data provided to the retry function.

Warning: The default retry count is 100!

Example "Pass instructions to the second run":

agent {
    name = "weather"
    filterOutput {
        if(!qualityCheck()) {
            retry(reason = "feedback to agent", max = 3)
        }
    }
    prompt {
        val retry = getOptional<RetrySignal>()
        if (retry != null) {
            // The details provided to the retry function can be accessed here.
            """ Updated Instructions with ${retry.reason}"""
        } else {
            """ Instructions """
        }
    }
}

In this example, a reason is passed to the retry. The reason can then be used to update the system prompt of the agent to help ensure a better response.