Grenadine Rasa Apa

Custom Slot Mappings#

The slots_mapped_in_domain argument provided to the required_slots method of FormValidationAction has been replaced by the domain_slots argument, please update your custom actions to the new argument name.

If none of the predefined Slot Mappings fit your use case, you can use the Custom Action validate_ to write your own extraction code. Rasa will trigger this action when the form is run.

If you're using the Rasa SDK we recommend you to extend the provided FormValidationAction. When using the FormValidationAction, three steps are required to extract customs slots:

In addition, you can override the required_slots method to add dynamically requested slots: you can read more in the Dynamic Form Behavior section.

If you have added a slot with a custom mapping in the slots section of the domain file which you only want to be validated within the context of a form by a custom action extending FormValidationAction, please make sure that this slot has a mapping of type custom and that the slot name is included in the form's required_slots.

The following example shows the implementation of a form which extracts a slot outdoor_seating in a custom way, in addition to the slots which use predefined mappings. The method extract_outdoor_seating sets the slot outdoor_seating based on whether the keyword outdoor was present in the last user message.

from typing import Dict, Text, List, Optional, Any

from rasa_sdk import Tracker

from rasa_sdk.executor import CollectingDispatcher

from rasa_sdk.forms import FormValidationAction

class ValidateRestaurantForm(FormValidationAction):

def name(self) -> Text:

return "validate_restaurant_form"

async def extract_outdoor_seating(

self, dispatcher: CollectingDispatcher, tracker: Tracker, domain: Dict

) -> Dict[Text, Any]:

text_of_last_user_message = tracker.latest_message.get("text")

sit_outside = "outdoor" in text_of_last_user_message

return {"outdoor_seating": sit_outside}

By default the FormValidationAction will automatically set the requested_slot to the first slot specified in required_slots which is not filled.

Custom Slot Mappings#

If none of the predefined Slot Mappings fit your use case, you can use the Custom Action validate_ to write your own extraction code. Rasa Open Source will trigger this action when the form is run.

If you're using the Rasa SDK we recommend you to extend the provided FormValidationAction. When using the FormValidationAction, three steps are required to extract customs slots:

The following example shows the implementation of a form which extracts a slot outdoor_seating in a custom way, in addition to the slots which use predefined mappings. The method extract_outdoor_seating sets the slot outdoor_seating based on whether the keyword outdoor was present in the last user message.

from typing import Dict, Text, List, Optional, Any

from rasa_sdk import Tracker

from rasa_sdk.executor import CollectingDispatcher

from rasa_sdk.forms import FormValidationAction

class ValidateRestaurantForm(FormValidationAction):

def name(self) -> Text:

return "validate_restaurant_form"

async def required_slots(

slots_mapped_in_domain: List[Text],

dispatcher: "CollectingDispatcher",

domain: "DomainDict",

) -> Optional[List[Text]]:

required_slots = slots_mapped_in_domain + ["outdoor_seating"]

return required_slots

async def extract_outdoor_seating(

self, dispatcher: CollectingDispatcher, tracker: Tracker, domain: Dict

) -> Dict[Text, Any]:

text_of_last_user_message = tracker.latest_message.get("text")

sit_outside = "outdoor" in text_of_last_user_message

return {"outdoor_seating": sit_outside}

By default the FormValidationAction will automatically set the requested_slot to the first slot specified in required_slots which is not filled.

Select which actions should receive domain#

You can control if an action should receive a domain or not.

To do this you must first enable selective domain in you endpoint configuration for action_endpoint in endpoints.yml.

url: "http://localhost:5055/webhook" # URL to your action server

enable_selective_domain: true

After selective domain for custom actions is enabled, domain will be sent only to those custom actions which have specifically stated that they need it. Custom actions inheriting from rasa-sdk FormValidationAction parent class are an exception to this rule as they will always have the domain sent to them. To specify if an action needs the domain add {send_domain: true} to custom action in the list of actions in domain.yml:

- action_hello_world: {send_domain: True} # will receive domain

- action_calculate_mass_of_sun # will not receive domain

- validate_my_form # will receive domain

Responses go under the responses key in your domain file or in a separate "responses.yml" file. Each response name should start with utter_. For example, you could add responses for greeting and saying goodbye under the response names utter_greet and utter_bye:

If you are using retrieval intents in your assistant, you also need to add responses for your assistant's replies to these intents:

utter_chitchat/ask_name:

- text: Oh yeah, I am called the retrieval bot.

utter_chitchat/ask_weather:

- text: Oh, it does look sunny right now in Berlin.

Notice the special format of response names for retrieval intents. Each name starts with utter_, followed by the retrieval intent's name (here chitchat) and finally a suffix specifying the different response keys (here ask_name and ask_weather). See the documentation for NLU training examples to learn more.

Select which actions should receive domain#

You can control if an action should receive a domain or not.

To do this you must first enable selective domain in you endpoint configuration for action_endpoint in endpoints.yml.

url: "http://localhost:5055/webhook" # URL to your action server

enable_selective_domain: true

After selective domain for custom actions is enabled, domain will be sent only to those custom actions which have specifically stated that they need it. Custom actions inheriting from rasa-sdk FormValidationAction parent class are an exception to this rule as they will always have the domain sent to them. To specify if an action needs the domain add {send_domain: true} to custom action in the list of actions in domain.yml:

- action_hello_world: {send_domain: True} # will receive domain

- action_calculate_mass_of_sun # will not receive domain

- validate_my_form # will receive domain

Test Conversation Format#

The test conversation format is a format that combines both NLU data and stories into a single file for evaluation. Read more about this format in Testing Your Assistant.

This format is only used for testing and cannot be used for training.

End-to-end training is an experimental feature. We introduce experimental features to get feedback from our community, so we encourage you to try it out! However, the functionality might be changed or removed in the future. If you have feedback (positive or negative) please share it with us on the Rasa Forum.

With end-to-end training, you do not have to deal with the specific intents of the messages that are extracted by the NLU pipeline or with separate utter_ responses in the domain file. Instead, you can include the text of the user messages and/or bot responses directly in your stories. See the training data format for detailed description of how to write end-to-end stories.

You can mix training data in the end-to-end format with labeled training data which has intents and actions specified: Stories can have some steps defined by intents/actions and other steps defined directly by user or bot utterances.

We call it end-to-end training because policies can consume and predict actual text. For end-to-end user inputs, intents classified by the NLU pipeline and extracted entities are ignored.

Only Rule Policy and TED Policy allow end-to-end training.

RulePolicy uses simple string matching during prediction. Namely, rules based on user text will only match if the user text strings inside your rules and input during prediction are identical.

TEDPolicy passes user text through an additional Neural Network to create hidden representations of the text. In order to obtain robust performance you need to provide enough training stories to capture a variety of user texts for any end-to-end dialogue turn.

Rasa policies are trained for next utterance selection. The only difference to creating utter_ response is how TEDPolicy featurizes bot utterances. In case of an utter_ action, TEDPolicy sees only the name of the action, while if you provide actual utterance using bot key, TEDPolicy will featurize it as textual input depending on the NLU configuration. This can help in case of similar utterances in slightly different situations. However, this can also make things harder to learn because the fact that different utterances have similar texts make it easier for TEDPolicy to confuse these utterances.

End-to-end training requires significantly more parameters in TEDPolicy. Therefore, training an end-to-end model might require significant computational resources depending on how many end-to-end turns you have in your stories.

To use forms with Rasa you need to make sure that the Rule Policy is added to your policy configuration. For example:

Define a form by adding it to the forms section in your domain. The name of the form is also the name of the action which you can use in stories or rules to handle form executions. You will need to specify a list of slot names to the mandatory required_slots key.

The following example form restaurant_form will fill the slot cuisine and slot num_people.

You can define a list of intents to ignore for the whole form under the ignored_intents key. Intents listed under ignored_intents will be added to the not_intent key of each slot mapping.

For example, if you do not want any of the required slots of a form to be filled when the intent is chitchat, then you would need to define the following (after the form name and under the ignored_intents keyword):

Once the form action gets called for the first time, the form gets activated and will prompt the user for the next required slot value. It does this by looking for a response called utter_ask__ or utter_ask_ if the former isn't found. Make sure to define these responses in your domain file for each required slot.

To activate a form you need to add a story or rule, which describes when the assistant should run the form. In the case a specific intent triggering a form, you can for example use the following rule:

- rule: Activate form

- intent: request_restaurant

- action: restaurant_form

- active_loop: restaurant_form

The active_loop: restaurant_form step indicates that the form should be activated after restaurant_form was run.

A form will automatically deactivate itself once all required slots are filled. You can describe your assistant's behavior for the end of a form with a rule or a story. If you don't add an applicable story or rule, the assistant will automatically listen for the next user message after the form is finished. The following example runs the utterances utter_submit and utter_slots_values as soon as the form your_form filled all required slots.

# Condition that form is active.

- active_loop: restaurant_form

# Form is deactivated

- action: restaurant_form

- requested_slot: null

# The actions we want to run when the form is submitted.

- action: utter_submit

- action: utter_slots_values

Users might want to break out of a form early. Please see Writing Stories / Rules for Unhappy Form Paths on how to write stories or rules for this case.

As of 3.0, slot mappings are defined in the slots section of the domain. This change allows the same slot mapping to be reused across multiple forms, removing any unnecessary duplication. Please follow the migration guide to update your assistant.

Note specifically the role of Mapping Conditions and the unique entity mapping constraint.

Calling Responses from Custom Actions#

You can use the responses to generate response messages from your custom actions. If you're using Rasa SDK as your action server, you can use the dispatcher to generate the response message, for example:

from rasa_sdk.interfaces import Action

class ActionGreet(Action):

return 'action_greet'

def run(self, dispatcher, tracker, domain):

dispatcher.utter_message(template="utter_greet")

If you use a different custom action server, your server should return the following JSON to call the utter_greet response:

"template":"utter_greet"

To use forms with Rasa Open Source you need to make sure that the Rule Policy is added to your policy configuration. For example:

Define a form by adding it to the forms section in your domain. The name of the form is also the name of the action which you can use in stories or rules to handle form executions. You also need to define slot mappings for each slot which your form should fill. You can specify one or more slot mappings for each slot to be filled.

The following example form restaurant_form will fill the slot cuisine from an extracted entity cuisine and slot num_people from entity number.

You can define a list of intents to ignore for the whole form under the ignored_intents key. Intents listed under ignored_intents will be added to the not_intent key of each slot mapping in the form.

For example, if you do not want any of the required slots of a form to be filled when the intent is chitchat, then you would need to define the following (after the form name and under the ignored_intents keyword):

The required_slots keyword was introduced. The following syntax is deprecated and will be removed in 3.0.0:

# this format is deprecated

Once the form action gets called for the first time, the form gets activated and will prompt the user for the next required slot value. It does this by looking for a response called utter_ask__ or utter_ask_ if the former isn't found. Make sure to define these responses in your domain file for each required slot.

To activate a form you need to add a story or rule, which describes when the assistant should run the form. In the case a specific intent triggering a form, you can for example use the following rule:

- rule: Activate form

- intent: request_restaurant

- action: restaurant_form

- active_loop: restaurant_form

The active_loop: restaurant_form step indicates that the form should be activated after restaurant_form was run.

A form will automatically deactivate itself once all required slots are filled. You can describe your assistant's behavior for the end of a form with a rule or a story. If you don't add an applicable story or rule, the assistant will automatically listen for the next user message after the form is finished. The following example runs the utterance utter_all_slots_filled as soon as the form your_form filled all required slots.

# Condition that form is active.

- active_loop: restaurant_form

# Form is deactivated

- action: restaurant_form

- requested_slot: null

# The actions we want to run when the form is submitted.

- action: utter_submit

- action: utter_slots_values

Users might want to break out of a form early. Please see Writing Stories / Rules for Unhappy Form Paths on how to write stories or rules for this case.

Rasa Open Source comes with four predefined mappings to fill the slots of a form based on the latest user message. Please see Custom Slot Mappings if you need a custom function to extract the required information.

The from_entity mapping fills slots based on extracted entities. It will look for an entity called entity_name to fill a slot slot_name. If intent_name is None, the slot will be filled regardless of intent name. Otherwise, the slot will only be filled if the user's intent is intent_name.

If role_name and/or group_name are provided, the role/group label of the entity also needs to match the given values. The slot mapping will not apply if the intent of the message is excluded_intent. Note that you can also define lists of intents for the parameters intent and not_intent.

not_intent: excluded_intent

In from_entity mapping, when an extracted entity uniquely maps onto a slot, the slot will be filled even if this slot wasn't requested by the form. The extracted entity will be ignored if the mapping is not unique.

In the example above, an entity date uniquely sets the slot arrival_date, an entity city with a role from uniquely sets the slot departure_city and an entity city with a role to uniquely sets the slot arrival_city, therefore they can be used to fit corresponding slots even if these slots were not requested. However, entity city without a role can fill both departure_city and arrival_city slots, depending which one is requested, so if an entity city is extracted when slot arrival_date is requested, it'll be ignored by the form.

The from_text mapping will use the text of the next user utterance to fill the slot slot_name. If intent_name is None, the slot will be filled regardless of intent name. Otherwise, the slot will only be filled if the user's intent is intent_name.

The slot mapping will not apply if the intent of the message is excluded_intent. Note that you can define lists of intents for the parameters intent and not_intent.

not_intent: excluded_intent

The from_intent mapping will fill slot slot_name with value my_value if user intent is intent_name or None. The slot mapping will not apply if the intent of the message is excluded_intent. Note that you can also define lists of intents for the parameters intent and not_intent.

The from_intent slot mapping will not apply during the initial activation of the form. To fill a slot based on the intent that activated the form, use the from_trigger_intent mapping.

not_intent: excluded_intent

The from_trigger_intent mapping will fill slot slot_name with value my_value if the form was activated by a user message with intent intent_name. The slot mapping will not apply if the intent of the message is excluded_intent. Note that you can also define lists of intents for the parameters intent and not_intent.

- type: from_trigger_intent

not_intent: excluded_intent

Session configuration#

A conversation session represents the dialogue between the assistant and the user. Conversation sessions can begin in three ways:

the user begins the conversation with the assistant,

the user sends their first message after a configurable period of inactivity, or

a manual session start is triggered with the /session_start intent message.

You can define the period of inactivity after which a new conversation session is triggered in the domain under the session_config key.

Available parameters are:

The default session configuration looks as follows:

session_expiration_time: 60 # value in minutes, 0 means infinitely long

carry_over_slots_to_new_session: true # set to false to forget slots between sessions

This means that if a user sends their first message after 60 minutes of inactivity, a new conversation session is triggered, and that any existing slots are carried over into the new session. Setting the value of session_expiration_time to 0 means that sessions will not end (note that the action_session_start action will still be triggered at the very beginning of conversations).

A session start triggers the default action action_session_start. Its default implementation moves all existing slots into the new session. Note that all conversations begin with an action_session_start. Overriding this action could for instance be used to initialize the tracker with slots from an external API call, or to start the conversation with a bot message. The docs on Customizing the session start action shows you how to do that.

Calling Responses as Actions#

If the name of the response starts with utter_, the response can directly be used as an action, without being listed in the actions section of your domain. You would add the response to the domain:

- text: "Hey! How are you?"

You can use that same response as an action in your stories:

- action: utter_greet

When the utter_greet action runs, it will send the message from the response back to the user.

If you want to change the text, or any other part of the response, you need to retrain the assistant before these changes will be picked up.

Conditional Response Variations#

Specific response variations can also be selected based on one or more slot values using a conditional response variation. A conditional response variation is defined in the domain or responses YAML files similarly to a standard response variation but with an additional condition key. This key specifies a list of slot name and value constraints.

When a response is triggered during a dialogue, the constraints of each conditional response variation are checked against the current dialogue state. If all constraint slot values are equal to the corresponding slot values of the current dialogue state, the response variation is eligible to be used by your conversational assistant.

The comparison of dialogue state slot values and constraint slot values is performed by the equality "==" operator which requires the type of slot values to match too. For example, if the constraint is specified as value: true, then the slot needs to be filled with a boolean true, not the string "true".

In the following example, we will define one conditional response variation with one constraint, that the logged_in slot is set to true:

influence_conversation: False

influence_conversation: False

text: "Hey, {name}. Nice to see you again! How are you?"

- text: "Welcome. How is your day going?"

- action: action_log_in

- action: utter_greet

In the example above, the first response variation ("Hey, {name}. Nice to see you again! How are you?") will be used whenever the utter_greet action is executed and the logged_in slot is set to true. The second variation, which has no condition, will be treated as the default and used whenever logged_in is not equal to true.

It is highly recommended to always provide a default response variation without a condition to guard against those cases when no conditional response matches filled slots.

During a dialogue, Rasa will choose from all conditional response variations whose constraints are satisfied. If there are multiple eligible conditional response variations, Rasa will pick one at random. For example, consider the following response:

text: "Hey, {name}. Nice to see you again! How are you?"

name: eligible_for_upgrade

text: "Welcome, {name}. Did you know you are eligible for a free upgrade?"

- text: "Welcome. How is your day going?"

If logged_in and eligible_for_upgrade are both set to true then both the first and second response variations are eligible to be used, and will be chosen by the conversational assistant with equal probability.

You can continue using channel-specific response variations alongside conditional response variations as shown in the example below.

influence_conversation: False

influence_conversation: False

text: "Hey, {name}. Nice to see you again on Slack! How are you?"

- text: "Welcome. How is your day going?"

Rasa will prioritize the selection of responses in the following order:

You can make responses rich by adding visual and interactive elements. There are several types of elements that are supported across many channels:

Here is an example of a response that uses buttons:

- text: "Hey! How are you?"

payload: "/mood_great"

Each button in the list of buttons should have two keys:

If you would like the buttons to also pass entities to the assistant:

- text: "Hey! Would you like to purchase motor or home insurance?"

- title: "Motor insurance"

payload: '/inform{{"insurance":"motor"}}'

- title: "Home insurance"

payload: '/inform{{"insurance":"home"}}'

Passing multiple entities is also possible with:

'/intent_name{{"entity_type_1":"entity_value_1", "entity_type_2": "entity_value_2"}}'

You can use buttons to overwrite the NLU prediction and trigger a specific intent and entities.

Messages starting with / are sent handled by the RegexInterpreter, which expects NLU input in a shortened /intent{entities} format. In the example above, if the user clicks a button, the user input will be classified as either the mood_great or mood_sad intent.

You can include entities with the intent to be passed to the RegexInterpreter using the following format:

/inform{"ORG":"Rasa", "GPE":"Germany"}

The RegexInterpreter will classify the message above with the intent inform and extract the entities Rasa and Germany which are of type ORG and GPE respectively.

You need to write the /intent{entities} shorthand response with double curly braces in domain.yml so that the assistant does not treat it as a variable in a response and interpolate the content within the curly braces.

Keep in mind that it is up to the implementation of the output channel how to display the defined buttons. For example, some channels have a limit on the number of buttons you can provide. Check your channel's documentation under Concepts > Channel Connectors for any channel-specific restrictions.

You can add images to a response by providing a URL to the image under the image key:

- text: "Here is something to cheer you up:"

image: "https://i.imgur.com/nGF1K8f.jpg"

Session configuration#

A conversation session represents the dialogue between the assistant and the user. Conversation sessions can begin in three ways:

the user begins the conversation with the assistant,

the user sends their first message after a configurable period of inactivity, or

a manual session start is triggered with the /session_start intent message.

You can define the period of inactivity after which a new conversation session is triggered in the domain under the session_config key.

Available parameters are:

The default session configuration looks as follows:

session_expiration_time: 60 # value in minutes, 0 means infinitely long

carry_over_slots_to_new_session: true # set to false to forget slots between sessions

This means that if a user sends their first message after 60 minutes of inactivity, a new conversation session is triggered, and that any existing slots are carried over into the new session. Setting the value of session_expiration_time to 0 means that sessions will not end (note that the action_session_start action will still be triggered at the very beginning of conversations).

A session start triggers the default action action_session_start. Its default implementation moves all existing slots into the new session. Note that all conversations begin with an action_session_start. Overriding this action could for instance be used to initialize the tracker with slots from an external API call, or to start the conversation with a bot message. The docs on Customizing the session start action shows you how to do that.

The config key in the domain file maintains the store_entities_as_slots parameter. This parameter is used only in the context of reading stories and turning them into trackers. If the parameter is set to True, this will result in slots being implicitly set from entities if applicable entities are present in the story. When an entity matches the from_entity slot mapping, store_entities_as_slots defines whether the entity value should be placed in that slot. Therefore, this parameter skips adding an explicit slot_was_set step manually in the story. By default, this behaviour is switched on.

You can turn off this functionality by setting the store_entities_as_slots parameter to false:

store_entities_as_slots: false

If you're looking for information on the config.yml file, check out the docs on Model Configuration.

A story is a representation of a conversation between a user and an AI assistant, converted into a specific format where user inputs are expressed as intents (and entities when necessary), while the assistant's responses and actions are expressed as action names.

Here's an example of a dialogue in the Rasa story format:

- story: collect restaurant booking info # name of the story - just for debugging

- intent: greet # user message with no entities

- action: utter_ask_howcanhelp

- intent: inform # user message with entities

- action: utter_on_it # action that the bot should execute

- action: utter_ask_cuisine

- action: utter_ask_num_people

While writing stories, you do not have to deal with the specific contents of the messages that the users send. Instead, you can take advantage of the output from the NLU pipeline, which lets you use just the combination of an intent and entities to refer to all the possible messages the users can send to mean the same thing.

It is important to include the entities here as well because the policies learn to predict the next action based on a combination of both the intent and entities (you can, however, change this behavior using the use_entities attribute).

All actions executed by the bot, including responses are listed in stories under the action key.

You can use a response from your domain as an action by listing it as one in a story. Similarly, you can indicate that a story should call a custom action by including the name of the custom action from the actions list in your domain.

During training, Rasa does not call the action server. This means that your assistant's dialogue management model doesn't know which events a custom action will return.

Because of this, events such as setting a slot or activating/deactivating a form have to be explicitly written out as part of the stories. For more info, see the documentation on Events.

Slot events are written under slot_was_set in a story. If this slot is set inside a custom action, add the slot_was_set event immediately following the custom action call. If your custom action resets a slot value to None, the corresponding event for that would look like this:

- story: set slot to none

# ... other story steps

- action: my_custom_action

There are three kinds of events that need to be kept in mind while dealing with forms in stories.

A form action event (e.g. - action: restaurant_form) is used in the beginning when first starting a form, and also while resuming the form action when the form is already active.

A form activation event (e.g. - active_loop: restaurant_form) is used right after the first form action event.

A form deactivation event (e.g. - active_loop: null), which is used to deactivate the form.

In order to get around the pitfall of forgetting to add events, the recommended way to write these stories is to use interactive learning.