Built-in NLP for Stealth bots via Microsoft's Conversational Language Understanding (CLU) This integration implements the Microsoft Conversational Language Understanding (CLU) service. It utilizes the built-in NLP features of Stealth 2.0.0.beta.
For instructions on creating your first Azure CLU Project, and on configuring your CLU project, please refer to the Analyze Conversation API documentation.
Once your account is set up, add these configuration settings to your services.yml file:
default: &default
nlp_integration: clu
clu:
subscription_key: <your-clu-subscription-key>
project_name: <your-clu-project>
deployment_name: <your-clu-deployment-name>
endpoint: <your-clu-endpoint>
production:
<<: *default
development:
<<: *default
test:
<<: *defaultUse the appropriate deployment_name to manage your different environment endpoints (e.g., development, staging, production).
That's it! Stealth will now automatically use CLU for intent detection and entity extraction automatically via handle_message and get_match methods.
Microsoft annonced that LUIS will be fully retired on October 1, 2025, and the LUIS portal will no longer be available starting October 31, 2025. It's recommended to migrate your LUIS applications to CLU to benefit from continued product support and multilingual capabilities, here are the instructions.
For now using stealth-clu for your bot:
- Switch your
Gemfileto usestealth-clu, instead ofstealth-luisand adjust configuration for your CLU project/endpoint in yourservices.ymlfile. - Entity names remain mostly the same. Some entities are now handled differently in CLU. For example,
Calendar.Durationfrom LUIS is now included underdatetimeV2with subtypeduration.
We recommend you name your intents using snake case (snake_case). This is because this integration will automatically convert your intent names to Ruby symbols.
So for example, if you have a handle_message defined like this:
handle_message(
'Maybe' => proc { step_to state: :say_maybe },
:yes => proc { step_to state: :say_yes },
:no => proc { step_to state: :say_no }
)If your user responds with a variation of the string maybe, then they will be taken to the state say_maybe.
Otherwise, the intent named yes and the intent named no will attempt to be matched. So if you had named your intent YES for example, you'd have to use :YES here which doesn't match Ruby syntax conventions.
The entity types listed below are named using their corresponding Stealth type. The equivalent type used by Microsoft CLU is also listed. For each code sample, the sample query is first provided followed by the array of entities extracted from the queries (for the given type).
It's possible, and even likely, that a query matches more than one entity type. For example, a currency type will also match a number type.
CLU prebuilt entity: number
"I think it was something like 63 or maybe 764"
[
63,
764
]
"It was almost 15k"
[
15000
]For more info about these values, please reference the number entity CLU documentation.
CLU prebuilt entity: money
"send me $87 or 48 cents"
[
{ 'number' => 87, 'units' => 'Dollar' },
{ 'number' => 48, 'units': 'Cent' }
]For more info about these values, please reference the currency entity CLU documentation.
CLU prebuilt entity: email
"you can contact me at john@email.none"
[
"john@email.none"
]For more info about these values, please reference the email entity CLU documentation.
CLU prebuilt entity: phonenumber
Note: CLU does not parse nor attempts to clean up phone number.
"You can reach me at 313-555-1212"
[
"313-555-1212"
]For more info about these values, please reference the phonenumber entity CLU documentation.
CLU prebuilt entity: percentage
"The stock is up 8.9% today"
[
8.9
]For more info about these values, please reference the percentage entity CLU documentation.
CLU prebuilt entity: age
"81 years old"
[
{ 'number' => 81, 'units' => 'Year' }
]For more info about these values, please reference the age entity CLU documentation.
CLU prebuilt entity: url
"please visit google.com or https://google.com"
[
"google.com",
"https://google.com"
]For more info about these values, please reference the url entity CLU documentation.
CLU prebuilt entity: ordinalV2
"they finished 2nd and 5th"
[
{ 'offset' => 2, 'relativeTo' => 'start' },
{ 'offset' => 5, 'relativeTo' => 'start' }
]
"she finished last"
[
{ 'offset' => 0, 'relativeTo' => 'end' }
]For more info about these values, please reference the ordinalV2 entity) CLU documentation.
CLU prebuilt entity: geographyV2
"She moved to paris, france"
[
{ 'value' => 'paris', 'type' => 'city' },
{ 'value' => 'france', 'type' => 'countryRegion' }
]For more info about these values, please reference the geographyV2 entity CLU documentation.
CLU prebuilt entity: dimension
"it's about 4 inches wide"
[
{ "number": 4, "units": "Inch" }
]For more info about these values, please reference the dimension entity CLU documentation.
CLU prebuilt entity: temperature
"it feels like 98 degrees"
[
{ 'number' => 98, 'units' => 'Degree' }
]For more info about these values, please reference the temperature entity CLU documentation.
CLU prebuilt entity: datetimeV2
This one is the most complicated one to work with. The values are nested pretty deeply. This integration exposes the values at such a high level because there is a chance that CLU will return results for more than one date type. For example, below we have just one result of type date, but CLU could return more than one object of subtype daterange, time, timerange, etc. See the docs for more info about these subtypes.
"tomorrow at 3pm"
[
{
"category" => "datetimeV2",
"text" => "tomorrow at 3pm",
"offset" => 9,
"length" => 15,
"confidenceScore" => 1,
"resolutions" => [
{
"resolutionKind" => "DateTimeResolution",
"dateTimeSubKind" => "DateTime",
"timex" => "2025-06-17T15",
"value" => "2025-06-17 15:00:00"
}
],
"extraInformation" => [
{
"extraInformationKind" => "EntitySubtype",
"value" => "datetime.datetime"
}
]
}
]For more info about these values, please reference the datetimeV2 entity CLU documentation.
CLU prebuilt entity: personName
"Little Cindy-Lou Who who was not more than two"
[
"Little Cindy-Lou"
]For more info about these values, please reference the personName entity CLU documentation.