This is a proof of concept of one way agents could communicate remotely. In the current design, the app must know the host and port of each remote agent it must be able to communicate with. This could in time be replaced with some kind of distributed directory (or just DNS).

Since I only implemented the Agent interface, there were a few issues found where the Agent type is used, but there are more requirements than just this type. Such as the following methods being required reset_consecutive_auto_reply_counter, reply_at_receive, clear_history

It also currently uses HTTP calls to send messages and uses Python's http.server module which definitely should not be used in production. It should be replaced with either a different server, or a different technology entirely. ZMQ, GPRC etc

Known issues:

• Termination does not work correctly
• RequestReply is hardcoded to true by the receiver (hacky support for "TERMINATE" added)

TODO:

• Separate sample out into separate example

receiver = Receiver(port=45553)

rajans_cal = ConversableAgent(
    "rajans_cal",
    llm_config={"config_list": config_list},
    system_message="..."
)
jacks_cal = RemoteAgent("jacks_cal", host="localhost", port=45554)

receiver.register_agent(rajans_cal)
receiver.start()

rajans_cal.initiate_chat(jacks_cal, message="...", clear_history=False)

Example of two remote agents coordinating to determine a common meeting time for two users:

First two turns: