Extending Rasa

Developing a custom Rasa NLU component, channel, or policy is straightforward.

In the project you created with botfront init you will find a rasa folder containing the material needed to build your Rasa image. When you run botfront up, the Rasa image used is built from this folder.

Example: custom NLU component

We provided you with an exemple of custom NLU component. Look for the ignore_sample_component.py file in the rasa folder. All this component does is adding a field called it_works to the NLU results. Let's use it:

1. Start your project botfront up and run botfront watch. This will watch the rasa and actions folders for changes and rebuild/restart the images when necessary so every change you make will be reloaded automatically.
2. Rename ignore_sample_component.py to sample_component.py (the watcher ignores files containing ignore). Botfront will rebuild and restart the Rasa image with your extensions installed as a pip package.
3. Go to NLU -> Settings -> Pipeline and add you new component:

pipeline:
  - name: sample_component.SampleComponent
  - name: WhitespaceTokenizer
  ...

4. Train
5. Head to the API tab and observe the result, it_works :)

{
  "intent": {
    "name": "chitchat.greet",
    "confidence": 0.9417732358
  },
  "entities": [],
  "language": "en",
  "it_works": true, // It just works
  "intent_ranking": [
    {
      "name": "chitchat.greet",
      "confidence": 0.9417732358
    },
    {
      "name": "chitchat.bye",
      "confidence": 0.0582267493
    }
  ],
  "text": "hello"
}

That's it. Creating a custom connector or core policy works exactly the same.