Response time improvement in phone interaction scenario

[Oceania2018]

Currently, if the Router is enabled, all user requests will go through the Router first, and the Router will decide the next action. Although this can enhance the processing power of AI, it will affect the response speed of AI and lead to a decrease in user experience. About 2 seconds will be added to the processing time for each request. This performance loss is especially obvious in phone interaction scenarios. Based on this consideration, I would like to add an option so that user requests will first enter the Agent. If the Agent believes that it cannot handle the user's new task request, then it will be transferred to the Router, and the Router will decide which Agent is more suitable to handle the current task.

In order to achieve this effect, the following changes need to be made:

1. Allow the Routing module to inject the fallback_to_router function into each Agent. If necessary, you may also need to add the corresponding system prompt, which may involve the prompt injection mechanism.
2. Add an optimized Agent selection capability to the Routing Service. In this function, whether the ability to transfer to manual customer service should be considered.
3. In fallback_to_route, call the GetSuitableAgent function in Routing Service to get the Agent Id, and then call InvokeAgent to get the reply to the current problem.

This method should be effective if the user clearly understands their tasks, especially in the phone IVR scenario.