When you publish a Databricks Mosaic AI connection in Application Integration, the Actions and Objects appear on the Metadata tab.
- •The Chat Completions action sends a chat completion request to a deployed model or a foundation model.
The request body contains a series of messages. The model generates a response to the last message, using earlier messages as context.
The following snippet is a request body example:
<chatCompletionRequest xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<messages>
<role>system</role>
<content>You are an AI that helps with complex problem solving.</content>
</messages>
<messages>
<role>user</role>
<content>Can you explain how quantum computing works?</content>
</messages>
<max_tokens m:type="xs:int">150</max_tokens>
<temperature m:type="xs:double">0.7</temperature>
<top_p m:type="xs:int">1</top_p>
<n m:type="xs:int">1</n>
</chatCompletionRequest>
- •The List Model Serving Endpoints action lists all the available model serving endpoints.
The following snippet is a sample response:
<root>
<endpoints>
<permission_level>CAN_MANAGE</permission_level>
<endpoint_type>FOUNDATION_MODEL_API</endpoint_type>
<task>llm/v1/chat</task>
<creation_timestamp>1699610000000</creation_timestamp>
<name>databricks-llama-4-maverick</name>
<last_updated_timestamp>1699610000000</last_updated_timestamp>
<state>
<config_update>NOT_UPDATING</config_update>
<ready>READY</ready>
</state>
<config>
<served_entities>
<name>databricks-llama-4-maverick</name>
<type>FOUNDATION_MODEL</type>
<foundation_model>
<input_price>7.143</input_price>
<price_unit>DBUs per 1M tokens</price_unit>
<docs>https://docs.databricks.com/machine-learning/foundation-models/supported-models.html#llama-4-maverick</docs>
<price>21.429</price>
<pricing_model>Pay-per-token</pricing_model>
<name>llama-4-maverick</name>
<description>Llama 4 Maverick is a state-of-the-art mixture of experts (MoE) language model trained and released by Meta. The model has 17B active parameters, 128 experts, and 400 billion total parameters. The model supports a context length of 128K tokens. The model is optimized for multilingual dialogue use cases, supporting 12 languages, and is aligned with human preferences for helpfulness and safety. It is not intended for use in languages other than English. Llama 4 is licensed under the Meta Llama 4 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. Customers are responsible for ensuring compliance with applicable model licenses.</description>
<display_name>Llama 4 Maverick</display_name>
</foundation_model>
</served_entities>
</config>
</endpoints>
</root>
- •The Get Genie Space action gets details of a Genie Space.
You must enter the space ID in the spaceID input field.
To find the space ID, perform the following steps:
- 1Click Genie.
- 2Select the space.
The following image shows the Genie space with the list of available spaces:
- 3Copy the space ID as shown in the following image:
- •The Start Conversation with Genie action starts a new conversation with the AI Genie.
You must enter the following input fields:
- - ConversationPrompt
<startConversationGenieRequest xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<content>Which product sold the most units?</content>
</startConversationGenieRequest>
- - spaceID
- •The List Vector Index action lists all the indexes at the given endpoint.
The following snippet is a sample request:
<root
xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<endpoint_name>docs-endpoint4</endpoint_name>
<delta_sync_index_spec>
<columns_to_sync>review</columns_to_sync>
<columns_to_sync>franchiseID</columns_to_sync>
<columns_to_sync>review_date</columns_to_sync>
<columns_to_sync>new_id</columns_to_sync>
<embedding_source_columns m:isArray="true">
<embedding_model_endpoint_name>databricks-bge-large-en</embedding_model_endpoint_name>
<name>review</name>
</embedding_source_columns>
<pipeline_type>TRIGGERED</pipeline_type>
<source_table>workspace.bakehouse2.media_customer_reviews</source_table>
</delta_sync_index_spec>
<index_type>DELTA_SYNC</index_type>
<name>workspace.bakehouse2.media_customer_reviews_indexes</name>
<primary_key>franchiseID</primary_key>
</root>
- •The List Vector Endpoints action lists all the vector endpoints.
- •The Create Vector Endpoint action creates a new vector endpoint.
The following snippet is a sample request:
<CreateVectorEndpoint xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<name>docs-endpoint4</name>
<endpoint_type>STANDARD</endpoint_type>
</CreateVectorEndpoint >
- •The Get Vector Endpoint action gets a vector endpoint.
- •The Create Vector Index action creates a new vector index.
The following snippet is a sample request:
<root
xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<endpoint_name>docs-endpoint4</endpoint_name>
<delta_sync_index_spec>
<columns_to_sync>review</columns_to_sync>
<columns_to_sync>franchiseID</columns_to_sync>
<columns_to_sync>review_date</columns_to_sync>
<columns_to_sync>new_id</columns_to_sync>
<embedding_source_columns m:isArray="true">
<embedding_model_endpoint_name>databricks-bge-large-en</embedding_model_endpoint_name>
<name>review</name>
</embedding_source_columns>
<pipeline_type>TRIGGERED</pipeline_type>
<source_table>workspace.bakehouse2.media_customer_reviews</source_table>
</delta_sync_index_spec>
<index_type>DELTA_SYNC</index_type>
<name>workspace.bakehouse2.media_customer_reviews_indexes</name>
<primary_key>franchiseID</primary_key>
</root>
- •The Query Vector Index action queries the specified vector index.
You must provide the following inputs:
- - indexName. For example: workspace.bakehouse2.media_customer_reviews_indexes
- - queryVectorIndexRequest
The following snippet is a sample request:
<root xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<num_results>5</num_results>
<columns>review</columns>
<columns>franchiseID</columns>
<query_text>Oatmeal Cookies</query_text>
</root>
- •The Create Embeddings action creates embeddings for the input text.
The following snippet is a sample request:
<root xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<input>Databricks</input>
</root>
- •The Update AI Gateway action updates the AI Gateway of a serving endpoint.
The following snippet is a sample request:
<updateAIGatewayRequest xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<usage_tracking_config>
<enabled>true</enabled>
</usage_tracking_config>
<rate_limits m:isArray="true">
<calls>15</calls>
<key>user</key>
<renewal_period>minute</renewal_period>
</rate_limits>
</updateAIGatewayRequest>
- •The Create Conversation Message action creates a new message in a conversation. The AI response uses all previously created messages in the conversation to respond.
You must enter the following input fields:
- - space_id
- - conversation_id
- - content
<root xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<content>What is the top 10 products sold?</content>
</root>
- •The Get Conversation Message action gets the message from the conversation.
You must enter the following input fields:
- - space_id
- - conversation_id
- - message_id
- •The Execute Message SQL action executes the SQL query for a message query attachment. Use this action when the query attachment has expired and needs to be executed again.
You must enter the following input fields:
- - space_id
- - conversation_id
- - message_id
- - attachment_id
- •The Get Message Attachment Result action gets the result of an SQL query if the message has a query attachment. This is only available if the message has a query attachment and the message status is EXECUTING_QUERY or COMPLETED.
You must enter the following input fields:
- - space_id
- - conversation_id
- - message_id
- - attachment_id
- •The Function Calling action provides an option to control the LLM outputs, ensuring that they generate structured responses more reliably.
The following snippet is a sample request:
<ChatCompletionRequest xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<model>databricks-llama-2-70b-chat</model>
<max_tokens m:type="xs:int">150</max_tokens>
<temperature m:type="xs:double">0.7</temperature>
<messages m:isArray="true">
<role>system</role>
<content>You are a helpful assistant. When asked for the age, you must use the available function.</content>
</messages>
<messages m:isArray="true">
<role>user</role>
<content>my name is john and I am 35 years old</content>
</messages>
<tools m:isArray="true">
<type>function</type>
<function>
<name>get_age_name</name>
<description>Get name and age of the person</description>
<parameters>
<type>object</type>
<properties>
<age>
<type>integer</type>
<description>The age of the person</description>
</age>
<name>
<type>string</type>
<description>The name of the person</description>
</name>
</properties>
<required m:isArray="true">age</required>
<required m:isArray="true">name</required>
</parameters>
</function>
</tools>
<tool_choice>auto</tool_choice>
</ChatCompletionRequest>
- •The Structured Output action provides an option to generate structured data in the form of JSON objects from the input data.
The following snippet is a sample request:
<root xmlns:m="urn:informatica:ae:xquery:json2xml:meta-data">
<max_tokens m:type="xs:int">150</max_tokens>
<temperature m:type="xs:double">0.7</temperature>
<messages m:isArray="true">
<role>system</role>
<content>You are an expert text analysis assistant. Your task is to extract specific entities from the provided text and present them using the available 'extract_entities' tool. Only use the tool provided. Do not add any conversational fluff before or after calling the tool.</content>
</messages>
<messages m:isArray="true">
<role>user</role>
<content>The contact person is Jane Doe, a Software Engineer at Acme Corp. You can reach her at jane.doe@acme.com for further inquiries.</content>
</messages>
<tools m:isArray="true">
<type>function</type>
<function>
<name>extract_entities</name>
<description>Extracts named entities like person name, company, role, and email from a given text.</description>
<parameters>
<type>object</type>
<properties>
<person_name>
<type>string</type>
<description>The full name of the person mentioned.</description>
</person_name>
<company_name>
<type>string</type>
<description>The name of the company mentioned.</description>
</company_name>
<role_title>
<type>string</type>
<description>The job title or role of the person.</description>
</role_title>
<contact_email>
<type>string</type>
<description>The email address mentioned.</description>
</contact_email>
</properties>
<required>person_name</required>
<required>company_name</required>
<required>role_title</required>
<required>contact_email</required>
</parameters>
</function>
</tools>
<tool_choice>
<type>function</type>
<function>
<name>extract_entities</name>
</function>
</tool_choice>
</root>