Warning: This document is for an old version of Rasa Core. The latest version is 0.14.5.

Slot Filling

One of the most common conversation patterns is to collect a few pieces of information from a user in order to do something (book a restaurant, call an API, search a database, etc.). This is also called slot filling.

If you need to collect multiple pieces of information in a row, we recommended that you create a FormAction. This is a single action which contains the logic to loop over the required slots and ask the user for this information. There is a full example using forms in the examples/formbot directory of Rasa Core.

You can take a look at the FormAction base class by clicking this link:

class rasa_core_sdk.forms.FormAction[source]

Domain Format

To add your forms to the domain file, reference their name under forms: section:

  - my_form

Forms Basics

Using a FormAction, you can describe all of the happy paths with a single story. By “happy path”, we mean that whenever you ask a user for some information, they respond with what you asked for.

If we take the example of the restaurant bot, this single story describes all of the happy paths.

## happy path
* request_restaurant
    - restaurant_form
    - form{"name": "restaurant_form"}
    - form{"name": null}

In this story the user intent is request_restaurant, which is followed by the form action restaurant_form. With form{"name": "restaurant_form"} the form is activated and with form{"name": null} the form is deactivated again. As shown in the section Handling unhappy paths the the bot can execute any kind of actions outside the form while the form is still active. On the “happy path”, where the user is cooperating well and the system understands the user input correctly, the form is filling all requested slots without interruption.

The FormAction will only requests slots which haven’t already been set. If a user says “I’d like a vegetarian Chinese restaurant for 8 people”, they won’t be asked about the cuisine and num_people slots.

Note that for this story to work, your slots should be unfeaturized. If they’re not, you should add all the slots that have been set by the form.

The restaurant_form in the story above is the name of our form action. Here is an example of what it looks like. You need to define three methods:

  • name: the name of this action
  • required_slots: a list of slots that need to be filled for the submit method to work.
  • submit: what to do at the end of the form, when all the slots have been filled.
class RestaurantForm(FormAction):
   """Example of a custom form action"""

   def name(self):
       # type: () -> Text
       """Unique identifier of the form"""

       return "restaurant_form"

   def required_slots(tracker):
       # type: () -> List[Text]
       """A list of required slots that the form has to fill"""

       return ["cuisine", "num_people", "outdoor_seating",
               "preferences", "feedback"]

   def submit(self, dispatcher, tracker, domain):
       # type: (CollectingDispatcher, Tracker, Dict[Text, Any]) -> List[Dict]
       """Define what the form has to do
           after all required slots are filled"""

       # utter submit template
       dispatcher.utter_template('utter_submit', tracker)
       return []

Once the form action gets called for the first time, the form gets activated and the FormPolicy jumps in. The FormPolicy is extremely simple and just always predicts the form action. See Handling unhappy paths for how to work with unexpected user input.

Every time the form action gets called, it will ask the user for the next slot in required_slots which is not already set. It does this by looking for a template called utter_ask_{slot_name}, so you need to define these in your domain file for each required slot.

Once all the slots are filled, the submit() method is called, where you can use the information you’ve collected to do something for the user, for example querying a restaurant API. If you don’t want your form to do anything at the end, just use return [] as your submit method. After the submit method is called, the form is deactivated, and other policies in your Core model will be used to predict the next action.

Custom slot mappings

Some slots (like cuisine) can be picked up using a single entity, but a FormAction can also support yes/no questions and free-text input. The slot_mappings method defines how to extract slot values from user responses.

Here’s an example for the restaurant bot:

    def slot_mappings(self):
        # type: () -> Dict[Text: Union[Dict, List[Dict]]]
        """A dictionary to map required slots to
            - an extracted entity
            - intent: value pairs
            - a whole message
            or a list of them, where a first match will be picked"""

        return {"cuisine": self.from_entity(entity="cuisine",
                "num_people": [self.from_entity(entity="num_people",
                "outdoor_seating": [self.from_entity(entity="seating"),
                "preferences": [self.from_intent(intent='deny',
                                                 value="no additional "
                "feedback": [self.from_entity(entity="feedback"),

The predefined functions work as follows:

  • self.from_entity(entity=entity_name, intent=intent_name) will look for an entity called entity_name to fill a slot slot_name regardless of user intent if intent_name is None else only if the users intent is intent_name.
  • self.from_intent(intent=intent_name, value=value) will fill slot slot_name with value if user intent is intent_name. To make a boolean slot, take a look at the definition of outdoor_seating above.
  • self.from_text(intent=intent_name) will use the next user utterance to fill the text slot slot_name regardless of user intent if intent_name is None else only if user intent is intent_name.
  • If you want to allow a combination of these, provide them as a list as in the example above

Validating user input

After extracting a slot value from user input, the form will try to validate the value of the slot. By default, this only checks if the requested slot was extracted. If you want to add custom validation, for example to check a value against a database, you can do this by writing validate_{slot}() method. Here is an example which checks if the extracted cuisine slot belongs to a list of supported cuisines.

    def cuisine_db():
        # type: () -> List[Text]
        """Database of supported cuisines"""
        return ["caribbean",
    def validate_cuisine(self,
                         value: Text,
                         dispatcher: CollectingDispatcher,
                         tracker: Tracker,
                         domain: Dict[Text, Any]) -> Optional[Text]:
        """Validate cuisine value."""

        if value.lower() in self.cuisine_db():
            # validation succeeded
            return value
            dispatcher.utter_template('utter_wrong_cuisine', tracker)
            # validation failed, set this slot to None, meaning the
            # user will be asked for the slot again
            return None

You can also deactivate the form directly during this validation step (in case the slot is filled with something that you are certain can’t be handled) by returning self.deactivate()

If nothing is extracted from the user’s utterance for any of the required slots, an ActionExecutionRejection error will be raised, meaning the action execution was rejected and therefore Core will fall back onto a different policy to predict another action.

Handling unhappy paths

Of course your users will not always respond with the information you ask of them. Typically, users will ask questions, make chitchat, change their mind, or otherwise stray from the happy path. The way this works with forms is that a form will raise an ActionExecutionRejection if the user didn’t provide the requested information. You need to handle events that might cause ActionExecutionRejection errors in your stories. For example, if you expect your users to chitchat with your bot, you could add a story like this:

## chitchat
* request_restaurant
    - restaurant_form
    - form{"name": "restaurant_form"}
* chitchat
    - utter_chitchat
    - restaurant_form
    - form{"name": null}

In some situations, users may change their mind in the middle of form action and decide not to go forward with their initial request. In cases like this, the assistant should stop asking for the requested slots. You can handle such situations gracefully using a default action action_deactivate_form which will deactivate the form and reset the requested slot. An example story of such conversation could look as follows:

## chitchat
* request_restaurant
    - restaurant_form
    - form{"name": "restaurant_form"}
* stop
    - utter_ask_continue
* deny
    - action_deactivate_form
    - form{"name": null}

It is strongly recommended that you build these stories using interactive learning. If you write these stories by hand you will likely miss important things. Please read Interactive Learning with Forms on how to use interactive learning with forms.

The requested_slot slot

The slot requested_slot is automatically added to the domain as an unfeaturized slot. If you want to make it featurized, you need to add it to your domain file as a categorical slot. You might want to do this if you want to handle your unhappy paths differently depending on what slot is currently being asked from the user. For example, say your users respond to one of the bot’s questions with another question, like why do you need to know that? The response to this explain intent depends on where we are in the story. In the restaurant case, your stories would look something like this:

## explain cuisine slot
* request_restaurant
    - restaurant_form
    - form{"name": "restaurant_form"}
    - slot{"requested_slot": "cuisine"}
* explain
    - utter_explain_cuisine
    - restaurant_form
    - slot{"cuisine": "greek"}
    ( ... all other slots the form set ... )
    - form{"name": null}

## explain num_people slot
* request_restaurant
    - restaurant_form
    - form{"name": "restaurant_form"}
    - slot{"requested_slot": "num_people"}
* explain
    - utter_explain_num_people
    - restaurant_form
    - slot{"cuisine": "greek"}
    ( ... all other slots the form set ... )
    - form{"name": null}

Again, is is strongly recommended that you use interactive learning to build these stories. Please read Interactive Learning with Forms on how to use interactive learning with forms.

Handling conditional slot logic

Many forms require more logic than just requesting a list of fields. For example, if someone requests greek as their cuisine, you may want to ask if they are looking for somewhere with outside seating.

You can achieve this by writing some logic into the required_slots() method, for example:

def required_slots(tracker):
   # type: () -> List[Text]
   """A list of required slots that the form has to fill"""

   if tracker.get_slot('cuisine') == 'greek':
     return ["cuisine", "num_people", "outdoor_seating",
             "preferences", "feedback"]
     return ["cuisine", "num_people",
             "preferences", "feedback"]

This mechanism is quite general and you can use it to build many different kinds of logic into your forms.


To use forms, make sure to include the FormPolicy in your policy configuration file. For example:

  - name: "FormPolicy"


The first thing to try is to run your bot with the debug flag, see Debugging for details. If you are just getting started, you probably only have a few hand-written stories. This is a great starting point, but you should give your bot to people to test as soon as possible. One of the guiding principles behind Rasa Core is:

Learning from real conversations is more important than designing hypothetical ones

So don’t try to cover every possibility in your hand-written stories before giving it to testers. Real user behavior will always surprise you!

Have questions or feedback?

We have a very active support community on Rasa Community Forum that is happy to help you with your questions. If you have any feedback for us or a specific suggestion for improving the docs, feel free to share it by creating an issue on Rasa Core GitHub repository.