Very often the Action Server needs to read data sources that are closely related to the training data in order to support actions. An example in the rasa demo chatbot, Sara, is the file
intent_description_mapping.csv. Here intents are mapped to textual representations so that the action that uses this file can make represent intents with text when generating an utterance to the user.
Of course this file has to be in sync with the conversation data. If we decide to add an intent we also have to update this file. Having two data sources to update can make deployment cumbersome. One needs to rebuild and deploy the docker image of the action server if the CSV file is shipped with the image.
My question is whether it was considered by the Rasa team to add custom field/keys to the domain, so that custom information can be added to the model and have it accessible by the action server. In doing so, one would not need to potentially update the action server image every time that the conversation data is updated. The action server would have data up to date by consuming the custom keys from the Domain.