Rasa Core SDK live stream!


(Juste) #1

Join me on the second Rasa livestream - let's talk Rasa Core SDK


@trust_level_0 Hey everyone!

Last week we released a new version of Rasa Core which includes a bunch of new features and changes - Rasa Core SDK is one of them. I would like to invite you to our second live twitch stream where we will cover the most important things regarding the SDK and migration from Rasa Core v.0.10 to v.0.11.

Please reply to this topic with all the questions you have regarding the Rasa Core SDK, or the new release of Rasa Core in general and I will make sure to answer them all during the stream with code examples where applicable!

The stream will happen on Rasa Twitch channel this Thursday, September 6, at 8:00 PM CEST.

Looking forward to your questions and also chatting to you all on Thursday!


(Matheus Matos) #2

Hi guys, I’m missing two things:

ConsoleInputChannel()
Agent.train_online()

Is it discontinued now? How can I run the bot on the Console?


(Adrian Humphrey) #3

I had this same problem, you must run the rasa_core.run script. You can go to the rasa core github page and go to examples folder. Go into the restaurant bot folder and open Makefile for an example.


(Juste) #4

(Frederik Ring) #5

Since custom actions now require a dedicated server in 0.11, I was wondering if I need to take care of starting / stopping that server myself when using any of the commands from rasa_core.run?

It works as intended, but I find it slightly cumbersome to keep “core” and “sdk” in sync when developing content locally. Is there any way to make rasa_core.run aware of the actions server?


(Akela Drissner) #6

@matheusmatos please read the migration guide, the ConsoleInputChannel has been removed, you don’t need to use the run script, but there’s a description there of how to import the function from the run script

@m90 at the moment you need to make sure you start the server yourself. Though Core does warn you when it’s not started when trying to execute an action


(Juste) #7

Don’t forget to tune in for the stream today and post your questions here! :slight_smile:


(Juste) #8

(Roman Telepathy Ai) #9

hi all,

so custom actions are ONLY possible through a dedicated server? no chance to use them as before?

Question to the actions server - is it ready to production server or is it simple werkzeug one?

thx & best, r


(Akela Drissner) #10

Yes exactly, you always need to run the action server. But yes it’s production ready


(Roman Telepathy Ai) #11

Hi Akela, thx for reply. one last question to clear all the confusions.

in domain.yml there should be the name of the action, in stories.md also; e.g. “my_custom_action”. During training, only the name is respected (cross checking between stories and domain).

But when running agent, actions server should accept the event with action name and find internally which class is responsible for the given action name.

could you please confirm? and/or correct me if needed…

thx and best, r


(Akela Drissner) #12

so the server loads up all the actions from your python file, and reads out the names from their name function. So that way it pulls out the requested action


(Pravinraj Vincent) #13

Hi Juste,

I have the bot version running from command line. Can you direct me how to run the bot in a website. If you can make a tutorial or document to show an example similar to your weather tutorial on slack, it would be great.


(Weilingfeng1996) #14

My problem is similar to the one above. I can perform normal training and conversations in the terminal, including slot fills and action responses. But I won’t use the http request to achieve the effect in the terminal, because I see rasa-based projects. The logic in httpserver doesn’t seem to use the logic in action.py, it seems to need to use a tracker. HttpApi is the command line content in the official documentation, which is difficult for newcomers to understand. If this tutorial is more informative, thank you.