Warning: This document is for an old version of Rasa Core.

Rasa Core without Python

Note

In this tutorial we will build a demo app using Rasa Core as a HTTP server. Although, you do not need to write any python code - you still need to understand basic concepts like domains, stories, and actions.

Example Code on GitHub

Goal

We will create a simple bot that doesn’t require you to write your custom action code in python, but rather any other language. To enable that, the process is a bit different for handling a message - as Rasa Core can not execute the actions internally but needs to wait for you to execute them.

Let’s start by creating a project folder:

mkdir remotebot && cd remotebot

After this tutorial, the folder structure should look like this:

remotebot/
├── data/
│   ├── stories.md              # dialogue training data
│   └── concert_messages.md     # nlu training data
├── concert_domain_remote.yml   # dialogue configuration
└── nlu_model_config.json       # nlu configuration

The first steps of creating a bot are very similar to other Rasa Core bots. But let’s go through each of them anyway - we will get to the HTTP interface at the end!

1. Define a Domain

The domain is similar to other examples, but does not contain utter templates. This is because you need to handle the connection to the input and output yourself (e.g. sending and receiving messages from facebook).

Here is an example domain for our remotebot, concert_domain_remote.yml:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
slots:
  concerts:
    type: list
  venues:
    type: list

intents:
  - greet
  - thankyou
  - goodbye
  - search_concerts
  - search_venues
  - compare_reviews

entities:
  - name

action_factory: remote

actions:
  - default
  - greet
  - goodbye
  - youarewelcome
  - search_concerts
  - search_venues
  - show_concert_reviews
  - show_venue_reviews

One important difference is action_factory: remote. This tells Rasa that it is not supposed to run the actions on its own.

Note

You can choose whatever action names you like. Rasa Core will not (and for that matter, can not) check their validity. Rasa will use these names to notify you about which action needs to be executed. So you need to be able to know what to do given an action name.

2. Define an interpreter

We are going to use Rasa NLU as an interpreter, so let’s create some intent examples in data/concert_messages.md:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
## intent:goodbye
- bye
- goodbye
- good bye
- stop
- end
- farewell
- Bye bye
- have a good one

## intent:thankyou
- thanks
- thank you
- thank you very much
- nice

## intent:search_concerts
- I am looking for a concert
- where can I find a concert
- Whats the next event

## intent:search_venues
- what is a good venue
- where is the best venue
- show me a nice venue
- how about a venue

## intent:compare_reviews
- how do these compare
- what are the reviews
- what do people say about them
- which one is better

## intent:greet
- hey
- howdy
- hey there
- hello
- hi
- good morning
- good evening
- dear sir

Furthermore, we need a configuration file nlu_model_config.json for the NLU model:

1
2
3
4
5
6
{
  "pipeline": "spacy_sklearn",
  "path" : "./models",
  "project": "nlu",
  "data" : "./data/concert_messages.md"
}

We can now train a NLU model using our examples (make sure to install Rasa NLU first as well as spaCy).

Let’s run

python -m rasa_nlu.train -c nlu_model_config.json --fixed_model_name current

to train our NLU model. A new directory models/nlu/current should have been created containing the NLU model.

Note

To gather more insights about the above configuration and Rasa NLU features head over to the Rasa NLU documentation.

3. Define stories

We need to add couple of stories as well to define the flow (so we can finally get to the interesting part of running in remote mode). Let’s put them into data/stories.md:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
## greet
* _greet
    - greet

## happy
* _thankyou
    - youarewelcome

## goodbye
* _goodbye
    - goodbye

## venue_search
* _search_venues
    - search_venues
    - slot{"venues": [{"name": "Big Arena", "reviews": 4.5}]}

## concert_search
* _search_concerts
    - search_concerts
    - slot{"concerts": [{"artist": "Foo Fighters", "reviews": 4.5}]}

## compare_reviews_venues
* _compare_reviews
    - show_venue_reviews

## compare_reviews_concerts
* _compare_reviews
    - show_concert_reviews

To train the dialogue model, run:

python -m rasa_core.train -s data/stories.md -d concert_domain_remote.yml -o models/dialogue

This will train the model and store it into models/dialogue.

4. Running the server

Now we can use our trained dialogue model and the previously created NLU model to run our bot in server mode:

python -m rasa_core.server -d models/dialogue -u models/nlu/current -o out.log

And there we have it! The bot is running and waiting for your HTTP requests.

5. Using the server

All communication is going to happen over a HTTP interface. You need to send a request to that interface to start a conversation as well as for the actions you have run on your end.

A detailed explanation including examples you can use with your running server can be found in Starting a conversation.