38 DBMS_CLOUD_AI

The DBMS_CLOUD_AI package with Select AI facilitates and configures the generation of SQL queries from natural language prompts to generate, run, and explain SQL statements. It also enables synthetic data generation, text summarization and translation, along with direct LLM chat.

38.1 Installing DBMS_CLOUD Packages

The DBMS_CLOUD family of packages is not pre-installed or configured with Oracle database. You need to manually install the DBMS_CLOUD packages and also configure users or roles to use this package.

For information on installation and configuration with Oracle Database 19c, Release 19.30 and above, see Using the DBMS_CLOUD Family of Packages.

For installing the DBMS_CLOUD packages and configuring users or roles for Oracle Database 19c, Releases prior to 19.30, see the MOS-NOTE with Doc ID 2748362.1. Note that earlier releases provide limited DBMS_CLOUD functionality only.

38.2 DBMS_CLOUD_AI Overview

Learn how you can use the DBMS_CLOUD_AI package.

Use the DBMS_CLOUD_AI package to create AI profiles and configure them for access to a Large Language Model (LLM). Set the AI profile in the current database user session to perform tasks, such as generating, running, and explaining SQL. In addition, you can use this package for natural language-based interactions including chatting with LLMs.

To learn more about supported platforms and LLMs, see:

About Select AI.

38.3 Summary of DBMS_CLOUD_AI Subprograms

This section covers the DBMS_CLOUD_AI subprograms and supporting parameter or attribute references provided with Oracle Database.

Subprogram	Description
CREATE_PROFILE Procedure	This procedure creates a new AI profile for translating natural language prompts to SQL statements.
Profile Attributes	Provides AI profile attributes that you can configure.
CLEAR_PROFILE Procedure	This procedure clears an active AI profile in the current session.
DISABLE_PROFILE Procedure	This procedure disables an AI profile in the current database.
DROP_PROFILE Procedure	This procedure drops an existing AI profile.
ENABLE_PROFILE Procedure	This procedure enables an AI profile to use in the current database.
GENERATE Function	This function enables you to use Select AI in a stateless manner with your existing profile.
GENERATE_SYNTHETIC_DATA Function	This function generates synthetic data.
GET_PROFILE Function	This function returns the profile name used in the current session.
SET_ATTRIBUTE Procedure	This procedure sets AI profile attributes.
SET_ATTRIBUTES Procedure	This procedure enables you to set one or more AI profile attributes by accepting attribute name and value pairs in JSON format.
SET_PROFILE Procedure	This procedure sets AI profile for the current database.
ENABLE_DATA_ACCESS Procedure	Use this procedure to enable sending data to your LLM.
DISABLE_DATA_ACCESS Procedure	Use this procedure to disable sending data to your LLM.
CREATE_CONVERSATION Procedure	This procedure helps you to create a conversation.
CREATE_CONVERSATION Function	This function helps you to create a conversation and use the conversation ID in other procedures.
CREATE_CONVERSATION Attributes	Use the conversation attributes to customize your conversations.
UPDATE_CONVERSATION Procedure	This procedure updates an existing procedure with user-specified parameters.
SET_CONVERSATION_ID Procedure	This procedure sets conversation support in the current session.
GET_CONVERSATION_ID Function	This procedure helps you to get the conversation_id parameter.
CLEAR_CONVERSATION_ID Procedure	This procedure helps you to clear any conversation_id set in the current session.
DELETE_CONVERSATION_PROMPT Procedure	This procedure deletes a particular prompt.
DROP_CONVERSATION Procedure	This procedure deletes an entire conversation and its metadata.
SUMMARIZE Function	This function summarizes your content based on the parameters.
SUMMARIZE Parameters	Use the summarize attributes to customize summary generation.
TRANSLATE Function	This function enables you to translate your text into the specified target_language.

38.3.1 CREATE_PROFILE Procedure

The procedure creates a new AI profile for translating natural language prompts to SQL statement.

Syntax

DBMS_CLOUD_AI.CREATE_PROFILE(
   profile_name        IN  VARCHAR2,
   attributes          IN  CLOB      DEFAULT NULL,
   status              IN  VARCHAR2  DEFAULT NULL,
   description         IN  CLOB      DEFAULT NULL
);

Parameters

Parameter	Description
profile_name	A name for the AI profile. The profile name must follow the naming rules of Oracle SQL identifier. Maximum length of profile name is 125 characters. This is a mandatory parameter.
attributes	Profile attributes in JSON format. See AI Profile Attributes for more details. The default value is NULL.
status	Status of the profile. The default value is enabled.
description	Description for the AI profile. The default value is NULL.

Example

BEGIN
    DBMS_CLOUD_AI.CREATE_PROFILE(
        profile_name    => 'OpenAI',
        attributes      => JSON_OBJECT(
                               'provider' value 'openai',
                               'credential_name' value 'openai_cred'
                           ),
        status          => 'enabled',
        description     => 'AI profile to use OpenAI for SQL translation'
    );
END;

To view profile attributes, see

• DBA_CLOUD_AI_PROFILE_ATTRIBUTES View

• USER_CLOUD_AI_PROFILE_ATTRIBUTES View

38.3.2 Profile Attributes

Attributes of an AI profile help to manage and configure the behavior of the AI profile. Some attributes are optional and have a default value.

Attributes

Attribute Name	Description
azure_deployment_name	Name of the Azure OpenAI Service deployed model. The name can only include alphanumeric characters, underscore character (_) and a hyphen (-) character. The name cannot end with an underscore (_) or a hyphen (-). To know how to get the azure_deployment_name, see Create and deploy an Azure OpenAI Service resource.
azure_embedding_deployment_name	Name of the Azure OpenAI deployed embedding model. The name can only include alphanumeric characters, underscore, and hyphen. The name can't start or end with a hyphen or underscore.
azure_resource_name	Name of the Azure OpenAI Service resource. The resource name can only include alphanumeric characters and hyphens, and can't start or end with a hyphen. To know how to get the azure_resource_name, see Create and deploy an Azure OpenAI Service resource.
case_sensitive_values	Specifies whether Select AI should instruct the LLM to generate case-insensitive SQL for columns containing string or text values. The supported values are: true: Indicates the LLM to generate SQL queries with case-sensitive string values. false: Indicates the LLM to generate SQL queries that ignore case for string values. The default value is false.
comments	Includes table and column comments in the metadata used for translating natural language prompts using AI. BOOLEAN data type is supported. The valid values are TRUE or FALSE for a string with VARCHAR2 data type. The values are not case sensitive.
constraints	Manages whether to include referential integrity constraints such as primary and foreign keys in the metadata sent to the LLM. The valid values are TRUE or FALSE. The default value is FALSE. The values are not case sensitive. TRUE: Includes referential constraints in the augmented prompt metadata that is sent to the LLM. FALSE: Does not include referential integrity constraints.
conversation	A VARCHAR2 attribute that indicates if conversation history is enabled for a profile. Valid values are true or false. The default value is false. The values are not case sensitive.
conversation_length	You can specify conversation_length. You can change the existing value to a suitable value for your use case. This is not a mandatory parameter. See CREATE_CONVERSATION Attributes for details.
credential_name	The name of the credential to access the AI provider APIs. Credential using bearer tokens can be created by using the provider name as the user name and bearer token as the password. Vault Secret credentials are also supported. Principal authentication, for example, Azure service principal, is also supported. For more information on how to configure it, see Use Azure Service Principal to Access Azure Resources. This is a mandatory attribute. See CREATE_CREDENTIAL Procedure.
enforce_object_list	Specifies whether to restrict the LLM to generate SQL that uses only tables covered by the object list. The supported values are: true: Enforces the use of only the tables specified in the object list. false: Allows the LLM to use tables accessible to the user. The values are not case sensitive. The default value is false.
max_tokens	Denotes the number of tokens to predict per generation. Default is 1024. See Tokens and Tokenizers for more details.
model	The name of the AI model being used to generate responses. See Select Your AI Provider and LLMs and specify the model name. Note: Cohere: Smaller, "light" models are faster, while larger models perform better. Custom models can also be supplied with their full ID. OCI Generative AI: The Chat Models are supported for all Select AI actions such as runsql, showsql, explainsql, narrate, and chat. Select AI supports pretrained models for OCI Generative AI. Custom models can also be supplied with their full OCIDs. If you are supplying OCID or oci_endpoint_id, be sure to provide either oci_runtimetype or oci_apiformat depending on the OCI Chat models. To know more about supported models in OCI Generative AI, see Pretrained Foundational Models in Generative AI. This parameter is not used for Azure as the model is determined when you create your deployment in the Azure OpenAI Service portal. AWS: Amazon Bedrock foundation models require access permissions through the Amazon Bedrock console. See the AWS documentation to get the modelID. OpenAI compatible: See the AI provider documentation for the supported model names.
object_list	Array of JSON objects specifying the owner and object names that are eligible for natural language translation to SQL. To include all objects of a given user, omit the "name" and only specify the "owner" key in the JSON object. The following types of objects can be used: tables views materialized views global temporary tables external tables synonyms on the above object types For translation natural language to SQL, the object name, object owner, object columns and comments are sent to the AI provider using HTTPS requests. Avoid specifying objects with sensitive object name, column names or comments in the object list. AI providers may have limit on the size of metadata allowed in translation requests. Consider limiting the list of objects suitable for the natural language prompts by your application users. Format: [ {"owner": "SH", "name": "SALES", {"owner": "TEST_USER"} ] External tables created using sync of OCI Data Catalog or AWS Glue can also be used the object list. This helps in managing metadata in central Data Catalogs and use the metadata directly for translating natural language prompts using AI.
oci_apiformat	Specifies the format in which the API expects data to be sent and received. Use this attribute to generate text responses. This attribute applies to OCI Generative AI Chat Models in a dedicated AI cluster. Specify this attribute when you specify a model OCID in the model attribute or provide an endpoint in the oci_endpoint_id attribute. Supported values are: COHERE GENERIC Note: Use this attribute for OCI Generative AI Chat Models
oci_compartment_id	Specifies the OCID of the compartment you are permitted to access when calling the OCI Generative AI service. The compartment ID can contain alphanumeric characters, hyphens and dots. The default is the compartment ID of the PDB.
oci_endpoint_id	This attributes indicates the endpoint OCID of the Oracle dedicated AI hosting cluster. The endpoint ID can contain alphanumeric characters, hyphens and dots. To find the endpoint OCID, see Getting an Endpoint's Details in Generative AI. When you want to use the Oracle dedicated AI cluster, you must provide the endpoint OCID of the hosting cluster. By default, the endpoint ID is empty and the model is on-demand on a shared infrastructure.
oci_runtimetype	This attribute indicates the runtime type of the provided model. This attribute is applicable to OCI Generate Text models in a dedicated AI cluster. Specify this attribute when you specify a model OCID in the model attribute or provide an endpoint in the oci_endpoint_id attribute. All permitted values can be found in OCI Generative AI runtimeType. See LlmInferenceRequest Reference. The supported values are: COHERE LLAMA Note: This attribute is deprecated. Use oci_apiformat.
provider	AI provider for the AI profile. Supported providers: openai cohere azure database oci google anthropic huggingface aws This is a mandatory attribute.
provider_endpoint	Specifies the API endpoint for OpenAI-compatible providers. This attribute is specific only to OpenAI-compatible providers. Specify provider_endpoint as an attribute in the DBMS_CLOUD_AI.CREATE_PROFILE procedure instead of the provider attribute. See Use OpenAI Compatible Providers to learn how to obtain the parameter. Example: api.fireworks.ai/inference
region	This attribute indicates the location of the Generative AI cluster that you want to use. The region can contain alphanumeric characters and hyphen characters. Note: The Oracle Generative AI cluster is available in Chicago, Frankfurt, London, and other selected regions. See Regions with Generative AIto learn more. The default region for AWS is us-east-1. The default region is us-chicago-1.
stop_tokens	The generated text will be terminated at the beginning of the earliest stop sequence. Sequence will be incorporated into the text. The attribute value must be a valid array of string values in JSON format. stop_tokens takes a JSON array as input. To learn more about stop tokens or stop sequences, see OpenAI or Cohere documentation.
temperature	Sampling from Generate Text models incorporates randomness, so that the same prompt may yield different outputs each time you hit "generate". Temperature is a non-negative float number used to tune the degree of randomness. Lower temperatures mean less random generations. See Temperature for more details. This parameter is applicable to all the supported service providers.
vector_index_name	Name of the vector index. The vector index name must follow the naming rules of Oracle SQL identifier. Maximum length of the vector store name is 125 characters.
source_language	The language of the input text sent to the provider for translation. Accepts either the full language name or the language code (language_code) supported by providers. The values are not case sensitive.
target_language	The language into which the provider translates the text. Accepts either the full language name or the language code (language_code) supported by providers. The values are not case sensitive.

The following example is using Cohere as the provider and displays custom profile attributes:

BEGIN
  DBMS_CLOUD_AI.CREATE_PROFILE(
     profile_name => 'COHERE',
     attributes =>
      '{"provider": "cohere",
        "credential_name": "COHERE_CRED",
        "object_list": [{"owner": "ADB_USER"}],
        "max_tokens":512,
        "stop_tokens": [";"],
        "model": "command-nightly",
        "temperature": 0.5,
        "comments": true,
	 "source_language": "en",
	 "target_language": "french"
       }');
END;
/

The following example shows custom profile attributes using OCI Generative AI:

BEGIN                                                                        
  DBMS_CLOUD_AI.CREATE_PROFILE(                                              
      profile_name => 'GENAI',                                                             
      attributes => '{"provider": "oci",                                                                   
        "credential_name": "GENAI_CRED",                                     
        "object_list": [{"owner": "SH", "name": "customers"},                
                        {"owner": "SH", "name": "countries"},                
                        {"owner": "SH", "name": "supplementary_demographics"},
                        {"owner": "SH", "name": "profits"},                  
                        {"owner": "SH", "name": "promotions"},               
                        {"owner": "SH", "name": "products"}],
        "oci_compartment_id": "ocid1.compartment.oc1...",
	"oci_endpoint_id": "ocid1.generativeaiendpoint.oc1.us-chicago-1....",
	"region": "us-chicago-1",
	"model": "cohere.command-light",
	"oci_apiformat": "COHERE"            
       }');                                                                  
END;                                                                         
/ 

38.3.3 CLEAR_PROFILE Procedure

This procedure clears any active AI profile set in the current session. Once you clear an active profile, you can no longer use SELECT AI on the SQL command line or use the set AI profile when calling DBMS_CLOUD_AI.GENERATE. This procedure does not drop the profile.

Syntax

DBMS_CLOUD_AI.CLEAR_PROFILE;

Example

   BEGIN
    DBMS_CLOUD_AI.CLEAR_PROFILE;
   END;
   /

38.3.4 DROP_PROFILE Procedure

The procedure drops an existing AI profile. If the profile does not exist, then the procedure throws an error.

Syntax

DBMS_CLOUD_AI.DROP_PROFILE(
       profile_name        IN   VARCHAR2,
       force               IN   BOOLEAN DEFAULT FALSE
 );

Parameters

Parameter	Description
profile_name	Name of the AI profile
force	If TRUE, then the procedure ignores errors if AI profile does not exist. The default value for this parameter is FALSE.

Example

BEGIN
     DBMS_CLOUD_AI.DROP_PROFILE(profile_name => 'OPENAI');
END;
/

Usage Notes

Use force to drop a profile and ignore errors if AI profile does not exist.

38.3.5 ENABLE_PROFILE Procedure

This procedure enables the AI profile that the user specifies. The procedure changes the status of the AI profile to ENABLED.

Syntax

DBMS_CLOUD_AI.ENABLE_PROFILE(
     profile_name         IN   VARCHAR2
 );

Parameters

Parameter	Description
profile_name	Name for the AI profile to enable This parameter is mandatory.

Example to Enable AI Profile

BEGIN
     DBMS_CLOUD_AI.ENABLE_PROFILE(
         profile_name    => 'OPENAI'
     );
END;
/

38.3.6 DISABLE_PROFILE Procedure

This procedure disables the AI profile in the current database. The status of the AI profile is changed to DISABLED by this procedure.

Syntax

DBMS_CLOUD_AI.DISABLE_PROFILE(
      profile_name  IN  VARCHAR2
);

Parameters

Parameter	Description
profile_name	Name for the AI profile. This parameter is mandatory.

Example

BEGIN
     DBMS_CLOUD_AI.DISABLE_PROFILE(
         profile_name    => 'OPENAI'
     );
END;
/

38.3.7 GET_PROFILE Function

This function returns the AI profile name set in the current session.

Syntax

DBMS_CLOUD_AI.GET_PROFILE
;

Example

This example shows how you can display the name of the profile in the current session.

   SELECT DBMS_CLOUD_AI.GET_PROFILE
   from DUAL;

38.3.8 SET_ATTRIBUTE Procedure

This procedure enables you to set AI profile attributes. It is overloaded to accept attribute values of various types.

Syntax

DBMS_CLOUD_AI.SET_ATTRIBUTE(
      profile_name         IN   VARCHAR2,
      attribute_name       IN   VARCHAR2,
      attribute_value      IN   {BOOLEAN|VARCHAR2}
);

DBMS_CLOUD_AI.SET_ATTRIBUTE(
      profile_name         IN   VARCHAR2,
      attribute_name       IN   VARCHAR2,
      attribute_value      IN   CLOB DEFAULT NULL
);

Parameters

Only the owner can set or modify the attributes of the AI profile. For a list of supported attributes, see Profile Attributes.

Parameter	Description
profile_name	Name of the AI profile for which you want to set the attributes. This parameter is mandatory.
attribute_name	Name of the AI profile attribute This parameter is mandatory.
attribute_value	Value of the profile attribute. Value can be BOOLEAN, CLOB, NUMBER or VARCHAR2 type. The default value is NULL.

Examples

BEGIN
 DBMS_CLOUD_AI.SET_ATTRIBUTE(
   profile_name    => 'OPENAI',
   attribute_name  => 'credential_name',
   attribute_value => 'OPENAI_CRED_NEW'
 );
END;
/

The following example accepts NUMBER type as the attribute_value.

BEGIN
 DBMS_CLOUD_AI.SET_ATTRIBUTE(
   profile_name    => 'OCI_PROFILE',
   attribute_name  => 'temperature',
   attribute_value => 0.5
 );
END;
/

The following example accepts BOOLEAN type as the attribute_value.

BEGIN
 DBMS_CLOUD_AI.SET_ATTRIBUTE(
   profile_name    => 'OCI_PROFILE',
   attribute_name  => 'comments',
   attribute_value => 'true'
 );
END;
/

The following example accepts VARCHAR2 type as the attribute_value.

BEGIN
 DBMS_CLOUD_AI.SET_ATTRIBUTE(
   profile_name    => 'OCI_PROFILE',
   attribute_name  => 'model',
   attribute_value => 'meta.llama-3.3-70b-instruct'
 );
END;
/

38.3.9 SET_ATTRIBUTES Procedure

This procedure enables you to set one or more AI profile attributes. The attribute name and value pairs are provided as a JSON document.

Syntax

DBMS_CLOUD_AI.SET_ATTRIBUTES(
        profile_name        IN   VARCHAR2,
        attributes          IN   CLOB
);

Parameters

Only the owner can set or modify the attributes of the AI profile. For a list of supported attributes, see Profile Attributes.

Parameter	Description
profile_name	Name of the AI profile for which you want to set the attributes. This parameter is mandatory.
attributes	AI profile attributes in JSON format. This parameter is mandatory.

Example

The following example accepts attributes in JSON format.

BEGIN
  DBMS_CLOUD_AI.SET_ATTRIBUTES(
    profile_name => 'OCI_PROFILE',
    attributes   => '{"temperature": 0.2,
                     "max_tokens": 3000}'
  );
END;
/

38.3.10 GENERATE Function

This function provides AI translation when using a stateless database connection. With your existing AI profile, you can use this function to perform the supported actions such as showsql, runsql, explainsql, narrate, summarize, and chat. The default action is showsql.

Overriding some or all of the profile attributes is also possible using this function.

Syntax

DBMS_CLOUD_AI.GENERATE(
    prompt            IN  CLOB,
    profile_name      IN  VARCHAR2 DEFAULT NULL,
    action            IN  VARCHAR2 DEFAULT NULL,
    attributes        IN  CLOB     DEFAULT NULL,
    params            IN  CLOB
) RETURN CLOB;

Parameters

Parameter	Description
prompt	Natural language prompt to translate using AI. The prompt can include SELECT AI <action> as the prefix. The action can also be supplied separately as an "action" parameter. The action supplied in the prompt overrides the action parameter. Default action is showsql. This parameter is mandatory.
profile_name	Name of the AI profile. This parameter is optional if an AI profile is already set in the session using DBMS_CLOUD_AI.SET_PROFILE. The default value is NULL. The following conditions apply: If a profile is set in the current session, the user may omit profile_name argument in the DBMS_CLOUD_AI.GENERATE function. If the profile_name argument is supplied in the DBMS_CLOUD_AI.GENERATE function, it overrides any value set in the session using the DBMS_CLOUD_AI.SET_PROFILE procedure. If there is no profile set in the session using the DBMS_CLOUD_AI.SET_PROFILE procedure, the profile_name argument must be supplied in the DBMS_CLOUD_AI.GENERATE function. Note: For Database Actions, you can either specify profile_name argument in DBMS_CLOUD_AI.GENERATE or you can run two steps as a PL/SQL script: DBMS_CLOUD_AI.SET_PROFILE and DBMS_CLOUD_AI.GENERATE. EXEC DBMS_CLOUD_AI.set_profile('OPENAI'); ----------------------------------------------- SELECT DBMS_CLOUD_AI.GENERATE(prompt => 'how many customers', profile_name => 'OPENAI', action => 'runsql') FROM dual; ----------------------------------------------- SELECT DBMS_CLOUD_AI.GENERATE(prompt => 'how many customers', profile_name => 'OPENAI', action => 'showsql') FROM dual; ----------------------------------------------- SELECT DBMS_CLOUD_AI.GENERATE(prompt => 'how many customers', profile_name => 'OPENAI', action => 'explainsql') FROM dual; ------------------------------------------------ SELECT DBMS_CLOUD_AI.GENERATE(prompt => 'how many customers', profile_name => 'OPENAI', action => 'narrate') FROM dual; ------------------------------------------- SELECT DBMS_CLOUD_AI.GENERATE(prompt => 'what is oracle autonomous database', profile_name => 'OPENAI', action => 'chat') FROM dual; See Executing SQL Statements in the Code Editor for more information.
action	Action for translating natural prompt using AI. The supported actions include: runsql(default) showsql explainsql narrate summarize translate chat Descriptions of actions are included in Use AI Keyword to Enter Prompts.
attributes	Override specific AI profile attributes by supplying attributes in JSON format. See Profile Attributes for more details.
params	Specify conversation parameter. See CREATE_CONVERSATION Attributes. You can specify only the following parameter: conversation_id: The default value is the conversation id set in the current session. This is not a mandatory parameter.

Examples

Example: Using the GENERATE Function for Select AI Actions

The following examples illustrate runsql, showsql, explainsql, narrate, summarize, translate, and chat actions that can be used with the DBMS_CLOUD_AI.GENERATE function.

See also Use AI Keyword to Enter Prompts for more details.

An example with runsql action is as follows:

SELECT DBMS_CLOUD_AI.GENERATE(prompt       => 'how many customers',
                              profile_name => 'OPENAI',
                              action       => 'runsql)
FROM dual;

An example with showsql action is as follows:

SELECT DBMS_CLOUD_AI.GENERATE(prompt       => 'how many customers',
                              profile_name => 'OPENAI',
                              action       => 'showsql')
FROM dual;

An example with explainsql action is as follows:

SELECT DBMS_CLOUD_AI.GENERATE(prompt       => 'how many customers',
                              profile_name => 'OPENAI',
                              action       => 'explainsql)
FROM dual;

An example with narrate action is as follows:

SELECT DBMS_CLOUD_AI.GENERATE(prompt       => 'how many customers',
                              profile_name => 'OPENAI',
                              action       => 'narrate')
FROM dual;

An example with chat action is as follows:

SELECT DBMS_CLOUD_AI.GENERATE(prompt       => 'what is oracle autonomous database',
                              profile_name => 'OPENAI',
                              action       => 'chat')
FROM dual;

An example with summarize action is as follows:

SELECT DBMS_CLOUD_AI.GENERATE(
                prompt => TO_CLOB(
                            DBMS_CLOUD.GET_OBJECT(
                             credential_name => 'STORE_CRED',
                             object_uri => 'https://objectstorage.ca-toronto-1.oraclecloud.com' ||
                                '/n/namespace-string/b/bucketname/o/data_folder/' ||
                                'summary/test_4000_words.txt')),
                profile_name => 'GENAI_LLAMA',
                action => 'SUMMARIZE')
from DUAL;

Result:

The music streaming industry, led by Spotify, has revolutionized the way
people consume music, with streaming accounting for 80% of the American
recording industry's revenue. However, this shift has also complicated the
lives of artists trying to survive in an on-demand, hyper-abundant present.
Spotify's business model, which pays royalties based on an artist's
popularity, has led to concerns about the fairness of the system, with some
artists earning little to no royalties. The company's dominance has also
changed the way people listen to music, with a focus on convenience and
personalized playlists. Liz Pelly's book, "Mood Machine: The Rise of Spotify
and the Costs of the Perfect Playlist," explores the impact of Spotify's rise
on the music industry and listeners, arguing that the platform's emphasis on
affect and mood has led to a homogenization of music and a loss of autonomy
for listeners. As the music industry continues to evolve, questions remain
about the future of music creation and consumption, and whether artists will
be able to thrive in a system that prioritizes convenience and profit over
artistic expression.

These examples show translate action:

The following examples shows using translate action in the prompt.

Note:

Your AI profile must specify target language.

SELECT DBMS_CLOUD_AI.GENERATE('select ai translate text to be translated')
          FROM dual;
   

The following example shows translate action supplied in the DBMS_CLOUD_AI.GENERATE function along with target_language and source_language. This example uses generative AI translation. The input text this is a document in English (source_language: "en") is translated into French (target_language: "fr").

DECLARE
         l_attributes  clob := '{"target_language": "fr", "source_language": "en"}';
         output clob;
      BEGIN
         output := DBMS_CLOUD_AI.GENERATE(
                        prompt            => 'this is a document',
                        profile_name      => 'oci_translate',
                        action            => 'translate',
                        attributes        => l_attributes
                     );

Using the GENERATE Function in a Procedure

You can use DBMS_CLOUD_AI.GENERATE in a procedure and run the function. The following example takes an ai_prompt, profile_name, and action as input parameters and calls DBMS_CLOUD_AI.GENERATE

create or replace FUNCTION call_select_ai (ai_prompt  IN VARCHAR2, 
                                           ai_profile IN VARCHAR2,
                                           ai_action  IN VARCHAR2) -- valid for 'chat', 'narrate', 'showsql'
                                           RETURN CLOB AS sai_resp clob;
BEGIN
  sai_resp := DBMS_CLOUD_AI.GENERATE(prompt       => ai_prompt,
                                     profile_name => ai_profile,
                                     action       => ai_action);  
  return(sai_resp);
END call_select_ai;

Example: Using the GENERATE Function for Conversations

The following example shows how to use DBMS_CLOUD_AI.GENERATE function in a conversation setting. This example assumes that a conversation has already been created.

SELECT DBMS_CLOUD_AI.GENERATE(
        prompt       =>  'What is the difference in weather between Seattle and San Francisco?',
        profile_name =>  'GENAI',
        action       =>  'CHAT',
        params       =>  '{"conversation_id":"30C9DB6E-EA4D-AFBA-E063-9C6D46644B92"}') AS RESPONSE;

Result:

RESPONSE
--------------------------------------------------------------------------------
Seattle and San Francisco, both located in the Pacific Northwest and Northern 
California respectively, experience a mild oceanic climate. However, there are 
some notable differences in their weather patterns:
 
1. **Temperature**: San Francisco tends to be slightly warmer than Seattle, 
especially during the summer months. San Francisco's average temperature ranges 
from 45?F (7?C) in winter to 67?F (19?C) in summer, while Seattle's average 
temperature ranges from 38?F (3?C) in winter to 64?F (18?C) in summer.
 
2. **Rainfall**: Seattle is known for its rainy reputation, with an average 
annual rainfall of around 37 inches (94 cm). San Francisco receives less rainfall, 
with an average of around 20 inches (51 cm) per year. However, San Francisco's 
rainfall is more concentrated during the winter months, while Seattle's rainfall 
is more evenly distributed throughout the year.
 
......

38.3.11 SET_PROFILE Procedure

This procedure sets AI profile for current session.

After setting an AI profile for the database session, any SQL statement with the prefix SELECT AI is considered a natural language prompt. Depending on the action the you specify with the AI prefix, a response is generated using AI. To use the AI prefix, see Examples of Using Select AI and Use AI Keyword to Enter Prompts. Optionally, it is possible to override the profile attributes or modify attributes by specifying them in JSON format. See SET_ATTRIBUTE Procedure for setting the attributes.

The AI profile can only be set for current session if the owner of the AI profile is the session user.

To set an AI profile for all sessions of a specific database user or all user sessions in the database, consider using a database event trigger for AFTER LOGON event on the specific user or the entire database. See CREATE TRIGGER Statement for more details.

Syntax

DBMS_CLOUD_AI.SET_PROFILE(
    profile_name      IN  VARCHAR2,
);

Parameters

Parameter	Description
profile_name	A name for the AI profile in the current session. This parameter is mandatory.

Example

   BEGIN
        DBMS_CLOUD_AI.SET_PROFILE(
          profile_name    => 'OPENAI'
        );
   END;
   /

38.3.12 CREATE_CONVERSATION Procedure

This procedure enables you to create a conversation and automatically set the conversation_id within the procedure.

Note:

If you are using DBMS_CLOUD_AI.CREATE_CONVERSATION procedure, you can skip setting the conversation_id as the procedure automatically sets it.

Syntax

DBMS_CLOUD_AI.CREATE_CONVERSATION(
  attributes            IN CLOB DEFAULT NULL
);

Parameters

Parameter	Description
attributes	Attributes for conversation in JSON format. See CREATE_CONVERSATION Attributes for more details. The default value is NULL.

Example

Example: Create Conversation

The following example shows creating a conversation without any customization.

EXEC DBMS_CLOUD_AI.CREATE_CONVERSATION;

Result:

PL/SQL procedure successfully completed.

Example: Create Conversation with Custom Attributes

The following example shows creating a conversation with custom parameters such as title, description, retention_days and conversation_length.

-- Create conversation with custom attributes
SELECT DBMS_CLOUD_AI.CREATE_CONVERSATION(
               attributes => '{"title":"Conversation 1",
                               "description":"this is a description",
                               "retention_days":5,
                               "conversation_length":5}')
     AS conversation_id FROM dual;

38.3.13 CREATE_CONVERSATION Function

This function creates a conversation and returns its conversation_id that can be used in other procedures or functions such as DBMS_CLOUD_AI.SET_CONVERSATION_ID and DBMS_CLOUD_AI.GENERATE.

Oracle recommends setting conversation_id to enable conversation. Alternately, you can set conversation_id in the DBMS_CLOUD_AI.GENERATE function.

Note:

If you are using DBMS_CLOUD_AI.CREATE_CONVERSATION procedure, you can skip setting the conversation_id as the procedure automatically sets it.

Syntax

DBMS_CLOUD_AI.CREATE_CONVERSATION(
  attributes            IN CLOB DEFAULT NULL
) RETURN VARCHAR2;

Parameters

Parameter	Description
attributes	Attributes for conversation in JSON format. See CREATE_CONVERSATION Attributes for more details. The default value is NULL.

Example

Example: Create Conversation

The following example shows using DBMS_CLOUD_AI.CREATE_CONVERSATION function to create a conversation without any customization.

SELECT DBMS_CLOUD_AI.CREATE_CONVERSATION FROM DUAL;

Result:

CREATE_CONVERSATION
------------------------------------
30C9DB6E-EA4D-AFBA-E063-9C6D46644B92

Example: Create Conversation with Custom Attributes

The following example shows using DBMS_CLOUD_AI.CREATE_CONVERSATION function to specify attributes such as title, retention_days and conversation_length.

SELECT DBMS_CLOUD_AI.CREATE_CONVERSATION(
				attributes => '{"title":"This is a test conversation",
                               "retention_days":7,
                               "conversation_length":20}') 
FROM DUAL;

38.3.14 CREATE_CONVERSATION Attributes

These attributes manage conversation context, including how long to retain it, how many prompts with responses to store or display, and metadata like title and description for reference. Some attributes are optional and have a default value.

Attributes

Attribute Name	Default Value	Description
title	New Conversation	The user-assigned name for the conversation. If not provided, Select AI will have the LLM generate one when the conversation is first used with a prompt.
description	NULL	Provides a user-defined description summarizing the purpose or context of the conversation. If it's not provided, the LLM generates one when the conversation is first used with a prompt and update it again on the 5th use to include more accurate and relevant information.
retention_days	7	Specify the number of days to retain the conversation history. This is stored in the database from its creation date. If you omit the value, the systems sets it to default value of 7. If you set it to 0, the system retains the conversation until you manually delete it using the DBMS_CLOUD_AI.DROP_CONVERSATION procedure or DBMS_CLOUD.DELETE_ALL_OPERATIONS('CONVERSATION').
conversation_length	NULL	Specify the number of recent prompts and responses to include with the current prompt. The maximum allowed value is 999. You can override this value by specifying the conversation_length parameter in the DBMS_CLOUD_AI.GENERATE function or by setting it in the AI profile using SELECT AI <ACTION> <PROMPT>. Apply the following precedence rules for conversation_length: Value from the attributes argument in DBMS_CLOUD_AI.GENERATE Value set in the conversation takes the next priority Value defined in the AI profile takes the last priority If none of them specify the conversation_length, then the default value is 10.

The following example shows you how you can customize conversation attributes in the DBMS_CLOUD_AI.CREATE_CONVERSATION procedure.

-- Create conversation with custom attributes
SELECT DBMS_CLOUD_AI.CREATE_CONVERSATION(
               attributes => '{"title":"Conversation 1",
                               "description":"this is a description",
                               "retention_days":5,
                               "conversation_length":5}')
     AS conversation_id FROM dual;

38.3.15 UPDATE_CONVERSATION Procedure

This procedure updates an existing conversation with a specified value of the conversation attributes.

Syntax

DBMS_CLOUD_AI.UPDATE_CONVERSATION(
    conversation_id    IN VARCHAR2,
    attributes         IN CLOB
);

Parameters

Parameter	Description
conversation_id	Unique number assigned to a conversation. This is a mandatory parameter.
attributes	Attributes for conversation in JSON format. See CREATE_CONVERSATION Attributes for more details.

Example

EXEC DBMS_CLOUD_AI.UPDATE_CONVERSATION(
conversation_id => '30C9DB6E-EA4E-AFBA-E063-9C6D46644B92', 
attributes => '{"retention_days":20, 
		"description":"This a sample description", 
		"title":"Sample title", 
		"conversation_length":20}');

Result:

PL/SQL procedure successfully completed.

38.3.16 SET_CONVERSATION_ID Procedure

This procedure sets the current conversation to the specified ID. Subsequent prompts include existing conversation prompts based on the conversation's configured attributes.

Syntax

DBMS_CLOUD_AI.SET_CONVERSATION_ID(
    conversation_id   IN VARCHAR2
);

Parameters

Parameter	Description
conversation_id	Unique number assigned to a conversation in the current session. This parameter is mandatory.

Example

EXEC DBMS_CLOUD_AI.SET_CONVERSATION_ID('30C9DB6E-EA4D-AFBA-E063-9C6D46644B92');
   

Result:

PL/SQL procedure successfully completed.

38.3.17 GET_CONVERSATION_ID Function

This function returns the conversation ID currently set in the session using either the DBMS_CLOUD_AI.SET_CONVERSATION_ID or DBMS_CLOUS_AI.CREATE_CONVERSATION procedure. If you did not set a conversation, the function returns NULL. If you drop the conversation, the system clears it in the session as well.See CLEAR_CONVERSATION_ID Procedure.

Syntax

DBMS_CLOUD_AI.GET_CONVERSATION_ID
RETURN VARCHAR2;

Example

This example displays the conversation ID set in the current session.

SELECT DBMS_CLOUD_AI.GET_CONVERSATION_ID;

Result:

--------------------------------------------------------------------------------
30C9DB6E-EA4F-AFBA-E063-9C6D46644B92

38.3.18 CLEAR_CONVERSATION_ID Procedure

This procedure clears a conversation ID set in the session to disable the conversation feature for SELECT AI <ACTION> <PROMPT>. If you did not set a conversation, the system does not raise any error.

Syntax

DBMS_CLOUD_AI.CLEAR_CONVERSATION_ID;

Example

This example demonstrates displaying the current conversation ID in the session, clearing the ID, and verifying the change.

-- A conversation id is set in the session
SELECT DBMS_CLOUD_AI.GET_CONVERSATION_ID FROM dual;
 
GET_CONVERSATION_ID
--------------------------------------------------------------------------------
3A88BFF0-1D7E-B3B8-E063-9C6D46640ECD
 
 
-- Clear the conversation id
EXEC DBMS_CLOUD_AI.CLEAR_CONVERSATION_ID;
 
PL/SQL procedure successfully completed.
 
 
-- The conversation id is removed from the session
SELECT DBMS_CLOUD_AI.GET_CONVERSATION_ID FROM dual;
 
GET_CONVERSATION_ID
--------------------------------------------------------------------------------

38.3.19 DELETE_CONVERSATION_PROMPT Procedure

The procedure removes a certain prompt from the conversation.

Syntax

DBMS_CLOUD_AI.DELETE_CONVERSATION_PROMPT(
    conversation_prompt_id  IN VARCHAR2,
    force                   IN BOOLEAN DEFAULT FALSE
);

Parameters

Parameter	Description
conversation_prompt_id	Unique number assigned to a prompt in a conversation. You can find the prompt ID by querying DBA/USER_CLOUD_AI_CONVERSATION_PROMPTS view. See DBMS_CLOUD_AI Views. This is a mandatory parameter.
force	If TRUE, then the procedure ignores errors if the conversation_prompt_id does not exist. The default value for this parameter is FALSE.

Example

EXEC DBMS_CLOUD_AI.DELETE_CONVERSATION_PROMPT('30C9DB6E-EA61-AFBA-E063-9C6D46644B92');

Result:

PL/SQL procedure successfully completed.

38.3.20 DROP_CONVERSATION Procedure

The procedure removes the conversation and all its associated prompts and the associated responses. Once dropped, the conversation_id becomes invalid. If a conversation is dropped while it's set in the session, it is cleared automatically.

Syntax

DBMS_CLOUD_AI.DROP_CONVERSATION(
    conversation_id  IN VARCHAR2,
    force            IN BOOLEAN  DEFAULT FALSE
);

Parameters

Parameter	Description
conversation_id	Unique number assigned to a conversation. This is a mandatory parameter.
force	If TRUE, then the procedure ignores errors if the conversation_id does not exist. The default value for this parameter is FALSE.

Example

EXEC DBMS_CLOUD_AI.DROP_CONVERSATION('30C9DB6E-EA4D-AFBA-E063-9C6D46644B92');

Result:

PL/SQL procedure successfully completed.

38.3.21 SUMMARIZE Function

This function summarizes your content based on the customization options you provide as parameters.

Syntax

DBMS_CLOUD_AI.SUMMARIZE(
  content         IN  CLOB     DEFAULT NULL,
  credential_name IN  VARCHAR2 DEFAULT NULL,
  location_uri    IN  VARCHAR2 DEFAULT NULL,
  profile_name    IN  VARCHAR2 DEFAULT NULL,
  user_prompt     IN  CLOB     DEFAULT NULL,
  params          IN  CLOB     DEFAULT NULL
) RETURN CLOB;

Parameters

Parameter	Description
content	Specifies the text you want to summarize. Either content or location_uri must be specified. This is not a mandatory parameter.
credential_name	Identifies the credential object used to authenticate with the object store. You must create this credential using DBMS_CLOUD.CREATE_CREDENTIAL. Note: Use this parameter only when you provide location_uri.
location_uri	Provides the URI where the text is stored or the path to a local file. Either content or location_uri must be specified. For example: Object storage: https://objectstorage.ca-toronto-1.oraclecloud.com/n/namespace-string/b/bucketname/o/data_folder/summary/test_file.txt Local file: summary_gobject:test_file.txt
profile_name	Specifies the AI profile to use. If you do not provide a profile_name, Select AI uses the default profile set in the session. If no default profile is set, it returns the error: ORA-20046: AI profile is not enabled in the session. The default value is NULL.
user_prompt	Supplies a natural language prompt to guide or customize the summary. You can include additional instructions beyond summary parameters. For example, The summary should start with ''The summary of the article is: ''' This parameter is not mandatory.
params	Defines summarization parameters. See SUMMARIZE Parameters.

Example

See Example: Select AI Summarize to explore.

38.3.22 SUMMARIZE Parameters

These attributes manage generating summary with custom parameters. Some attributes are optional and have a default value.

Attributes

Attribute Name	Default Value	Description
min_words	0	Specifies the approximate minimum number of words the generated summary is expected to contain. Note: This parameter acts as a guideline rather than a strict limit: the actual length of the summary may vary depending on the content provided and the model's interpretation.
max_words	200	Specifies the approximate maximum number of words the generated summary is expected to contain. Note: This parameter acts as a guideline rather than a strict limit, the actual length of the summary may vary depending on the content provided and the model's interpretation.
summary_style	Paragraph	Specifies the format style for the summary. The following are the available summary format options: paragraph: the summary is presented in one or more paragraphs. list: the summary is a list of key points from the text.
chunk_processing_method	map_reduce	When the text exceeds the token limit that the LLM can process, it must be split into manageable chunks. This parameter enables you to choose the method for processing these chunks. The following options are provided: iterative_refinement. See Iterative Refinement for more details. map_reduce. See Map Reduce for more details.
extractiveness_level	low	Determines how closely the summary follows the original wording of the input. It controls the degree to which the model extracts versus rephrases it. The following are the options: High: The summary stays close to the original phrasing, reusing sentences and expressions where possible. Medium: A balanced mix of extraction and paraphrasing. Low: Allows more freedom to reword, restructure, or simplify the input, focusing on meaning rather than exact phrasing. Note: This setting serves as guidance for the model’s summarization behavior, it does not enforce a strict rule. The actual style and wording of the summary may vary based on the input content and model decisions.

38.3.23 TRANSLATE Function

This function enables you to translate your text into the specified target_language.

You can supply the source_language and the target_language parameters in the function or they can be taken from the user's AI profile. If your AI profile does not include a source_language attribute, the generative AI provider automatically detects the input language. If the target_language attribute is missing, Select AI returns an error.

Syntax

DBMS_CLOUD_AI.TRANSLATE(
   profile_name      IN VARCHAR2,
   text              IN CLOB,
   source_language   IN VARCHAR2 DEFAULT NULL,
   target_language   IN VARCHAR2 DEFAULT NULL
) RETURN CLOB;

Parameters

Parameter	Description
profile_name	Specifies the AI profile to use. This is not a mandatory parameter.
text	Specifies the text you want to translate. This is a mandatory parameter.
source_language	Language of the input text
target_language	Language into which the text is translated.

Example

See Example: Select AI Translate to explore.

38.3.24 GENERATE_SYNTHETIC_DATA Function

Use this procedure to generate synthetic data for a single table, multiple tables or a full schema.

The following is the syntax to generate synthetic data for a single table.

Syntax

DBMS_CLOUD_AI.GENERATE_SYNTHETIC_DATA(
  profile_name        IN  VARCHAR2,
  object_name         IN  DBMS_ID,
  owner_name          IN  DBMS_ID,
  record_count        IN  NUMBER,
  user_prompt         IN  CLOB DEFAULT NULL,
  params              IN  CLOB DEFAULT NULL
);

The following is the syntax to generate synthetic data for multiple tables.

DBMS_CLOUD_AI.GENERATE_SYNTHETIC_DATA(
  profile_name        IN  VARCHAR2,
  object_list         IN  CLOB,
  params              IN  CLOB DEFAULT NULL
);

If you do not want table data or vector search documents to be sent to an LLM, a user with administrator privileges can disable such access for all users of the given database. This, in effect, disables the narrate action.

Parameters

Parameter	Mandatory	Description
profile_name	Yes	The AI profile containing necessary LLM service information. This can be created by CREATE_PROFILE Procedure.
object_name	Yes	Specify a table name to populate synthetic data. SELECT and INSERT privilege on the table objects are needed for the user using it. The table is either empty or have records in it.
owner_name	No	Specify the database user who owns the referenced object. If no specific owner is provided, the procedure defaults to using the schema of the user running it.
record_count	No	The number of records to be synthetically generated.
user_prompt	No	Additional information that a user can mention to generate synthetic data. For example, to generate a record for a table called MOVIE with a release_date column, the user_prompt can be: the release date for the movies should be in 2019
params	No	Optional attributes provided in JSON object string format to modify the behavior of an API. See Optional Parameters.
object_list	Yes	Use this parameter for generating synthetic data on multiple tables. This parameter takes in table object information along with it's arguments and contains the same arguments provided in the single table. See object_list Parameters.

Optional Parameters

Parameter	Value Datatype	Value	Description
sample_rows	Number	0 <= sample_rows <= 100	Specify the number of rows from the table to use as a sample to guide the LLM in data generation. A value of 0 means no sample rows will be used. The default value is 0.
table_statistics	Boolean	True False	Enable or disable the use of table statistics information. The default value is False.
priority	String	Valid values: HIGH MEDIUM LOW	Assign a priority value that defines the number of parallel requests sent to the LLM for generating synthetic data. Tasks with a higher priority will consume more database resources and complete faster. The default value is HIGH HIGH: Specifies the number of parallel LLM requests based on the database's ECPU count (or OCPU count if your database uses OCPUs). MEDIUM: Sets the number of concurrent processes according to the concurrency limit for Medium service. The default value is 4. LOW: Runs the pipeline job in a serial order, without parallel processing. The maximum number of concurrent parallel processes used for synthetic data generation is limited to 64.
comments	Boolean	True False	Enable or disable sending comments to the LLM to guide data generation. The default value is False.

object_list Parameters

Parameter	Value Datatype	Mandatory	Description
owner	String	Yes	Specifies the database user who owns the object being referenced. If no specific owner is provided, the procedure will default to using the schema of the user running it.
name	String	No	Specify a table name to populate synthetic data. SELECT and INSERT privilege on the table objects are needed for the user using it.The table is either empty or have records in it.
record_count	Number	No	The number of records to be synthetically generated. Provide a number greater than 0. Supply record_count or record_count_percentage.
record_count_percentage	Number	No	The percentage of number of records to be synthetically generated. Provide a number greater than 0. For a Metadata Clone database, where the table metadata including statistics is preserved, the record_count_percentage parameter is supported. Supply record_count or record_count_percentage. When using the record_count_percentage parameter, the final record count in the table is calculated as: Original_Num_Rows * record_count_percentage
user_prompt	String	No	Same as user_prompt in Parameters. The user_prompt is associated with a specific table object.

Examples

The following examples show the DBMS_CLOUD_AI.GENERATE_SYNTHETIC_DATA function for generating synthetic data for a single table and multiple tables. For a complete example and to view more examples, see Example: Generate Synthetic Data.

BEGIN
    DBMS_CLOUD_AI.GENERATE_SYNTHETIC_DATA(
        profile_name => 'GENAI',
        object_name  => 'Director',
        owner_name   => 'ADB_USER',
        record_count => 5
    );
END;
/
PL/SQL procedure successfully completed.

BEGIN
    DBMS_CLOUD_AI.GENERATE_SYNTHETIC_DATA(
        profile_name => 'GENAI',
        object_list => '[{"owner": "ADB_USER", "name": "Director","record_count":5},
                         {"owner": "ADB_USER", "name": "Movie_Actor","record_count":5},
                         {"owner": "ADB_USER", "name": "Actor","record_count":10},
                         {"owner": "ADB_USER", "name": "Movie","record_count":5,"user_prompt":"all movies are released in 2009"}]'
    );
END;
/
PL/SQL procedure successfully completed.

38.3.25 ENABLE_DATA_ACCESS Procedure

This procedure enables sending data to LLM for applicable Select AI features, which is the default behavior. Only an administrator can run this procedure.

This procedure controls data access for the following Select AI capabilities:

• narrate action
• Retrieval Augmented Generation (RAG)
• Synthetic Data Generation

Syntax

DBMS_CLOUD_AI.ENABLE_DATA_ACCESS();

Parameters

This procedure does not require any parameters.

Example to Enable Data Access

BEGIN
  DBMS_CLOUD_AI.ENABLE_DATA_ACCESS();
END;
/

38.3.26 DISABLE_DATA_ACCESS Procedure

This procedure disables sending data to LLM for applicable Select AI features. Only an administrator can run this procedure.

This procedure limits the following Select AI capabilities:

• narrate action
• Retrieval Augmented Generation (RAG)
• Synthetic Data Generation

Syntax

DBMS_CLOUD_AI.DISABLE_DATA_ACCESS();

Parameters

This procedure does not require any parameters.

Example to Disable Data Access

BEGIN
  DBMS_CLOUD_AI.DISABLE_DATA_ACCESS();
END;
/