Rasa

Rasa is an open source NLU provider that is distributed as an Python API, developed and provided by Rasa Technologies. Different from other closed source providers like Dialogflow, you will need to host your own server in order to use Rasa. This will pose several challenges and is not as easy to integrate into the framework as other agents, where you simply have to provide a valid configuration to the configuration file and are good to go. One major advantage of hosting your own server is that you do not have to share any of your data with another company. Here we will quickly introduce how you can setup emubot using Rasa.

Setting Up a Rasa Agent

You can install the Rasa API as a pip package by using the following command:

pip3 install rasa-x --extra-index-url https://pypi.rasa.com/simple

After you have installed the Rasa API you can follow Rasa tutorial to define your own Rasa bot. After you have done so you can use the provided Rasa commands to train your bot and start a test server:

rasa train
rasa run --enable-api -m path/to/trained/model

The example Rasa server will start on port 5000 by default. The framework uses the HTTP API provided with the server as a reference for its implementation. If you want to use your own server between the framework and Rasa, you should provide the following two routes for the requests from the framework:

POST /model/parse

This route parses the provided text using your model.

{
    message_id: string,
    sender: 'user',
    text: string,
    token: string
}

The response is expected to include the following fields:

{
    entities: [
        {
            value: string,
            entity: string,
        }
    ],
    intent: {
        confidence: number,
        name: string
    },
}

Where entities and intent are conform with the Rasa parse results.

You can also add the following fields in order to give additional information to your interceptors:

{
    entities: [
        {
            value: string,
            entity: string,
        }
    ],
    intent: {
        confidence: number,
        name: string
    },
    action: string,
    contexts: [string]
}

POST /webhooks/rest/webhook

This route parses the provided text with your trained Rasa model.

{
    message: string,
    message_id: string,
    token: string,
}

The result of the request is the answer of the Rasa agent:

{
    "text": string
}

Note that emubot will also add the token you provide in the configuration to the requests. You can use this token as an API token for your server.

Of course you can include any number of other information in your responses, which you can then use to trigger events in the implementation of your Interceptors. But you should make sure to pass the information to the emubot framework.

These requests have to conform to the definition in the Rasa HTTP API. Head over to their docs for more information.

Exemplar Rasa Configuration

In this section you can see an example for a Rasa configuration to be used in the emubot framework.

export const platformNlpRasa = {
    agents: {
        first_agent: {
            defaultLifespan: 2,
            executionIndex: 1,
            languageCode: 'en',
            minScore: 0.75,
            token: env.NLP_TOKEN,
            url: 'https://my-cool-rasa-bot.com/',
        },
    },
    constructor: RasaAdapter,
    name: 'rasa',
};