How can the server distinguish between the client's invocation methods of send_message and send_message_streaming?

[Linux2010]

In the HelloWorld program of the example, both client.send_message(request) and client.send_message_streaming(streaming_request) will execute the HelloWorldAgentExecutor.execute method. So how can we distinguish between these two invocation methods within the execute method? It seems there is no attribute in the current context to identify the difference.

(In fact, the current HelloWorld program does not well demonstrate the characteristics of streaming.) I think if we can distinguish the caller, I can call the agent's invoke method and stream method separately in the execute method.

[iangoldby]

You don't need to - from the point of view of the agent, the SDK abstracts away the distinction between streaming and non-streaming. Just write the agent as if it were a streaming agent and the SDK will ensure it works properly with non-streaming clients too.

Just call TaskUpdater.update_status() from your agent executor execute() method each time the agent itself yields another piece of the response. If the client is non-streaming then it will get only the final task result. If it is streaming then it will incrementally get each individual update.