I'm creating a connector for Rasa X (rasa 1.0) to work with Google Assistant as a front-end. Before 1.0 came out, this tutorial: https://medium.com/rasa-blog/going-beyond-hey-google-building-a-rasa-powered-google-assistant-5ff916409a25 worked really well. However, when I tried to run Rasa X on the same structure there were incompatibilities between my old Flask-based connector and the Rasa agent starting the project.
In the new version, rasa.core.agent.handle_channels([input_channel], http_port=xxxx)
uses Sanic instead, which seems to be incompatible with my old method.
I have tried to convert the old Flask connector to a Sanic(never used it before) one and I used Postman to check the health route, which works. I also receive the payload from Assistant. However, when I forward this to the Rasa agent, I get nothing in return.
- This is the new Sanic connector:
class GoogleAssistant(InputChannel):
@classmethod
def name(cls):
return "google_assistant"
def blueprint(self, on_new_message):
# this is a Sanic Blueprint
google_webhook = sBlueprint("google_webhook")
@google_webhook.route("/", methods=['GET'])
def health(request):
return response.json({"status": "ok"})
@google_webhook.route("/webhook", methods=["POST"])
def receive(request):
#payload = json.loads(request)
payload = request.json
sender_id = payload["user"]['userId']
intent = payload['inputs'][0]['intent']
text = payload['inputs'][0]['rawInputs'][0]['query']
try:
if intent == "actions.intent.MAIN":
message = "<speak>Hello! <break time=\"1\"/> Welcome to the Rasa-powered Google Assistant skill. You can start by saying hi."
else:
# Here seems to be the issue. responses is always empty
out = CollectingOutputChannel()
on_new_message(UserMessage(text,out,sender_id))
responses = [m["text"] for m in out.messages]
message = responses[0]
except LookupError as e:
message = "RASA_NO_REPLY"
print(e)
r = json.dumps("some-response-json")
return HTTPResponse(body=r, content_type="application/json")
return google_webhook
- And this is the script starting the project:
action_endpoint = EndpointConfig(url="http://localhost:5055/webhook")
nlu_interpreter = RasaNLUInterpreter('path-to-model')
agent = Agent.load('path-to-model-2', interpreter = nlu_interpreter, action_endpoint=action_endpoint)
input_channel = GoogleAssistant()
agent.handle_channels([input_channel], http_port=5010)
I expect the output to be the text selected by the Rasa agent as a reply "This is the reply", but I get nothing (the list is empty).
EDIT
I defined def receive(request):
as async def receive(request):
and changed on_new_message(UserMessage(text,out,sender_id))
to await on_new_message(UserMessage(text,out,sender_id))
. Furthermore, the script starting the project is now:
loop = asyncio.get_event_loop()
action_endpoint = EndpointConfig(url="http://localhost:5055/webhook")
nlu_interpreter = RasaNLUInterpreter('path')
agent = Agent.load('path2', interpreter = nlu_interpreter, action_endpoint=action_endpoint)
input_channel = GoogleAssistant()
loop.run_until_complete(agent.handle_channels([input_channel], http_port=5010))
loop.close()
Unfortunately it didn't change anything, still not getting any reply from Rasa on the output channel.