multi-skill-assistants.rst

desc:	Iterate quickly by developing reusable building blocks of AI assistant skills and combining them at training time.

Multi-skill Assistants

.. edit-link::

You can build a contextual AI assistant by combining reusable "building blocks" called skills. You might, for example, handle chitchat with one skill and greet your users with another. These skills can be developed in isolation, and then combined at train time to create your assistant.

An example directory structure could look like this:

.
├── config.yml
└── skills
    ├── GreetBot
    │   ├── data
    │   │   ├── nlu.md
    │   │   └── stories.md
    │   └── domain.yml
    └── ChitchatBot
        ├── config.yml
        ├── data
        │   ├── nlu.md
        │   └── stories.md
        └── domain.yml

In this example the contextual AI assistant imports the ChitchatBot skill which in turn imports the GreetBot skill. Skill imports are defined in the configuration files of each project. To instruct Rasa to use the the module to import skills, put this section in the config file of your root project:

importers:
- name: SkillSelector

Then specify which skills you want to import. In our example, the config.yml in the root project would look like this:

imports:
- skills/ChitchatBot

The configuration file of the ChitchatBot in turn references the GreetBot:

imports:
- ../GreetBot

The GreetBot skill does not specify further skills so the config.yml can be omitted.

Rasa uses relative paths from the referencing configuration file to import skills. These can be anywhere on your file system as long as the file access is permitted.

During the training process Rasa will import all required training files, combine them, and train a unified AI assistant.

Note

Rasa will use the policy and NLU pipeline configuration of the root project directory during the training. Policy or NLU configurations of imported skills will be ignored.

Note

Equal identifiers will be merged, e.g. if two skills have training data for an intent greet, their training data will be combined.