README.md

Rasa Middleware Connector

Adds support for middlewares to Rasa connectors. The middlewares run before the message is sent to the Rasa agent allowing text/UserMessage preprocessing.

Why not add a message_preprocessor when sending the message to the Rasa Agent?

You can do it, and you can also add middleware support to the preprocessor.

This module tries to bring more functionality to pre-processing, allowing access to all fields in a UserMessage object and the possibility to cancel a message execution when preprocessing it.

When you set up a default message preprocessor by passing it as a callable argument on the function handle_message (usually called on_new_message on the channels), and it can even call a sequence of middlewares, but it only allows access to the text field on a UserMessage.

Usage

Input Connector

The class InputMiddlewareConnector is an abstract class that you should inherit from, together with the desired connector. It should be the first class.

class SocketInput(InputMiddlewareConnector, SocketIOInput):

You need to implement the methods: get_middlewares, get_on_new_message and create_user_message. And change the connector handle_message route to point to the self.handle_message. All the following examples are extracts from a custom connector based on the SocketIOInput:

def blueprint(self, on_new_message):

    # save Agent handler on object for later reference
    self.on_new_message = on_new_message

    # get the webhook from the super blueprint
    socketio_webhook = super().blueprint(on_new_message)
    self.sio = socketio_webhook.sio

    # swaps default event handler with the desired handler
    self.sio.handlers = socketio_webhook.sio.handlers
    self.sio.handlers['/'][self.user_message_evt] = self.handle_message

    return socketio_webhook

get_on_new_message: this method returns the Rasa Agent handle_message endpoint, usually it is received as the parameter on_new_message on the blueprint function of a handler. (On the example above we save this on self.on_new_message)

get_middlewares: this method should return a list with your middleware objects (created inheriting from BaseMiddleware class) in order of execution. The Rasa Agent handler will be added after the middlewares.

def get_middlewares(self):
    
    return [
        TextCleaner(),
        MessageCollector(self)        
    ]

create_user_message(self, *args, **kwargs): this method returns a UserMessage object containing your message.

def create_user_message(self, sid, data) -> UserMessage:

    # this function receives required data to create a new
    # user message and returns it (based on the official rasa
    # implementation of the socket connector) 
    
    if self.session_persistence:
        if not data.get("session_id"):
            logger.warning(
                "A message without a valid sender_id "
                "was received. This message will be "
                "ignored. Make sure to set a proper "
                "session id using the "
                "`session_request` socketIO event."
            )
            return
        sender_id = data["session_id"]
    else:
        sender_id = sid

    text = data["message"]

    output_channel = SocketIOOutput(self.sio, sender_id, self.bot_message_evt)
    message = UserMessage(
        text, output_channel, sid, input_channel=self.name()
    )
    
    return message

Output Connector

An output middleware connector is very similar to the input one. You should override get_middlewares just like on the input.

One caveat is that OutputMiddlewareConnector should be the first on the inheritance order. Also, there is the need to provide the connector class on get_connector_class:

class SocketOutput(OutputMiddlewareConnector, SocketIOOutput):
...
    def get_connector_class(self):
        return SocketIOOutput

Middlewares

Your middlewares should respect the interface defined in the class BaseMiddleware:

• set_next(self, next, is_output)
• compute(self, message: UserMessage)
• input_compute(self, message: UserMessage)
• output_compute(self, recipient_id: Text, message: Dict[Text, Any])
• Attributes: next, is_output

Note the at least input_compute or output_compute should be implemented. If you will only use the middleware as an input middleware you can only implement input_compute, and if its an only output middleware only output_compute needs to be implemented.

set_next: sets where your middleware should send the message after you are done processing it.

compute(self, message: UserMessage): starts the processing of your message. Sends it to the appropriate function (input_compute or output_compute).

input_compute or output_compute: message processing. After processing the message you should call:

• await self.next(message) if input
• await self.next(recipient_id, message) if output

Example:

async def input_compute(self, message: UserMessage):

    logger.info("Middleware TextCleaner received message from {}".format(message.sender_id))
    
    message.text = self.clean_message(message.text)
    await self.next(message)

# output_compute not implemented since this is an input only middleware