DBMS_CLOUD_AI Package

The DBMS_CLOUD_AI package, with Select AI, facilitates and configures the translation of natural language prompts to generate, run, explain SQL statements. Also, enables retrieval augmented generation and natural language-based interactions, including chatting with LLMs.

Prerequisites

An outbound connectivity must have been configured using a NAT gateway, by your fleet administrator as described below:
  • Create a NAT gateway in the Virtual Cloud Network (VCN) where your Autonomous AI Database resources reside by following the instructions in Create a NAT Gateway in Oracle Cloud Infrastructure Documentation.
  • After creating the NAT gateway, add a route rule and an egress security rule to each subnet (in the VCN) where Autonomous AI Database resources reside so that these resources can use the gateway to obtain a public key from your Azure AD instance:
    1. Go to the Subnet Details page for the subnet.
    2. In the Subnet Information tab, click the name of the subnet's Route Table to display its Route Table Details page.
    3. In the table of existing Route Rules, check whether there is already a rule with the following characteristics:
      • Destination: 0.0.0.0/0
      • Target Type: NAT Gateway
      • Target: The name of the NAT gateway you just created in the VCN

      If such a rule does not exist, click Add Route Rules and add a route rule with these characteristics.

    4. Return to the Subnet Details page for the subnet.
    5. In the subnet's Security Lists table, click the name of the subnet's security list to display its Security List Details page.
    6. In the side menu, under Resources, click Egress Rules.
    7. In the table of existing Egress Rules, check whether there is already a rule with the following characteristics:
      • Destination Type: CIDR
      • Destination: 0.0.0.0/0
      • IP Protocol: TCP
      • Source Port Range: 443
      • Destination Port Range: All

      If such a rule does not exist, click Add Egress Rules and add an egress rule with these characteristics.

The HTTP Proxy settings in your environment must allow the database to access the cloud service provider.

These settings are defined by your fleet administrator while creating the Exadata Cloud@Customer infrastructure as described in Using the Console to Provision Exadata Database Service on Cloud@Customer .

Note:

The network configuration including the HTTP Proxy can only be edited until the Exadata Infrastructure is in Requires Activation state. Once it is activated, you cannot edit those settings.

Setting up an HTTP Proxy for an already provisioned Exadata Infrastructure needs a Service Request (SR) in My Oracle Support. See Create a Service Request in My Oracle Support for details.

Related Topics

Summary of DBMS_CLOUD_AI Subprograms

This section covers the DBMS_CLOUD_AI subprograms provided with Autonomous AI 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.

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.

DISABLE_PROFILE Procedure

This procedure disables an AI profile in the current database.

SET_ATTRIBUTE Procedure

This procedure sets AI profile attributes.

SET_PROFILE Procedure

This procedure sets AI profile for the current database.

GENERATE Function

 This function generates a SQL statement using AI to translate.

GENERATE_SYNTHETIC_DATA Function

This function generates synthetic data.
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_VECTOR_INDEX Procedure

This procedure creates a vector index in the specified vector database, and populates it with data from an object store using an asynchronous scheduler job.

DROP_VECTOR_INDEX Procedure

This procedure removes a vector store index. It normally removes the vector store index object and deletes the vector database.

DISABLE_VECTOR_INDEX Procedure

This procedure disables a vector index object in the current database. When disabled, an AI profile cannot use the vector index, and the system does not load data into the vector store.

ENABLE_VECTOR_INDEX Procedure

This procedure enables or activates a previously disabled vector index object.

UPDATE_VECTOR_INDEX Procedure

This procedure updates an existing vector store index with a specified value of the vector index attribute.

Vector Index Attributes

Provides vector index profile attributes that you can configure.
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.
FEEDBACK Procedure

Use this procedure to potentially improve query generation accuracy by providing a feedback to Select AI.
Vector Index for FEEDBACK

This is a default vector index created when you first use feedback.

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 enable.

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'),
          description     => 'AI profile to use OpenAI for SQL translation'
     );
END;
/

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_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.

comments

Include 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.

Note:

Boolean values are not applicable in the DBMS_CLOUD_AI.SET_ATTRIBUTE procedure when setting a single attribute because attribute_value parameter is of CLOB datatype.

conversation

A VARCHAR2 attribute that indicates if conversation history is enabled for a profile. Only OpenAI and Azure OpenAI Service support conversation history. Valid values are true or false. The default value is false. The values are not case sensitive.

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.

This is a mandatory attribute. See CREATE_CREDENTIAL Procedure.

embedding_model

The embedding model defined in the AI profile. The following are the supported AI providers for the embedding models with their default values:

  • OCI GenAI: cohere.embed-english-v3.0
  • OpenAI: text-embedding-ada-002
  • Azure OpenAI: text-embedding-ada-002
  • Cohere: embed-english-v2.0
  • Google: text-embedding-004

Note:

The embedding_model parameter is not applicable for Anthropic and Hugging Face.

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.

Supported models for:

  • OpenAI: gpt-4, gpt-4-0613, gpt-4-32k, gpt-4-32k-0613, gpt-3.5-turbo (default), gpt-3.5-turbo-0613, gpt-3.5-turbo-16k, gpt-3.5-turbo-16k-0613
  • Cohere: command (default), command-nightly (experimental), command-light, and command-light-nightly (experimental). Smaller, "light" models are faster, while larger models perform better. Custom models can also be supplied with their full ID.

  • OCI Generative AI: For OCI Generative AI: cohere.command (default). Pretrained models for OCI Generative AI are all supported by Select AI. Custom models can also be supplied with their full OCIDs.

    To know more about supported models in OCI Generative AI, see Pretrained Foundational Models in Generative AI.

  • AWS: Amazon Bedrock foundation models require access permissions through the Amazon Bedrock console. See the AWS documentation to get the modelID.

Note:

This parameter is not used for Azure as the model is determined when you create your deployment in the Azure OpenAI Service portal.

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.

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"}
]

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 Autonomous AI Database.

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 required when the model attribute is specified and is applicable to OCI Generation Models.

All permitted values can be found in OCI Generative AI runtimeType. See LlmInferenceRequest Reference.

The supported values are:
  • COHERE
  • LLAMA

provider

AI provider for the AI profile.

Supported providers:

  • openai
  • cohere
  • azure
  • oci
  • google
  • anthropic
  • huggingface
  • aws

This is a mandatory attribute.

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, and London regions. See Pretrained Foundational Models in Generative AI.
The default region is us-chicago-1.

The default region for AWS is us-east-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.

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
       }');
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_runtimetype": "COHERE"            
       }');                                                                  
END;                                                                         
/

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.

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;
/

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;
/

SET_ATTRIBUTE Procedure

This procedure enables you to set AI profile attributes.

Syntax

DBMS_CLOUD_AI.SET_ATTRIBUTE(
      profile_name         IN   VARCHAR2,
      attribute_name       IN   VARCHAR2,
      attribute_value      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.

attribute_name

Name of the AI profile attribute

This parameter is mandatory.

attribute_value

Value of the profile attribute.

The default value is NULL.

Example

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

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 specified with the AI prefix, a response is generated using AI. To use the AI prefix, see 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 in Oracle Database 19c Database PL/SQL Language Reference or Oracle Database 26ai Database PL/SQL Language Reference 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;
   /

GENERATE Function

This function provides AI translation in a stateless manner. With your existing AI profile, you can use this function to perform the supported actions such as showsql, narrate, or 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
) 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 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       => 'showsql')
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 showsql (default), narrate, and chat. Descriptions of actions are included in Use AI Keywords to Enter Prompts.

Note:

This function does not support the runsql action. If you supply the runsql action, it returns the following error:
ORA-20000: runsql action is not supported by generate function ORA-06512: at "C##CLOUD$SERVICE.DBMS_CLOUD", line xxxx 
ORA-06512: at "C##CLOUD$SERVICE.DBMS_CLOUD_AI", line 2696 ORA-06512: at line x
attributes Override specific AI profile attributes by supplying attributes in JSON format. See Profile Attributes for more details.

Examples

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

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 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;

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 . 

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.

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;
/

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;
/

CREATE_VECTOR_INDEX Procedure

This procedure creates a vector index in the specified vector database, and populates it with data from an object store using an asynchronous scheduler job.

Syntax

PROCEDURE CREATE_VECTOR_INDEX(                                              
   index_name          IN  VARCHAR2,                                        
   attributes          IN  CLOB      DEFAULT NULL,                          
   status              IN  VARCHAR2  DEFAULT NULL,                          
   description         IN  CLOB      DEFAULT NULL                           
);

Parameters

Parameter Description

index_name

Name of the vector index. The vector index name must follow the naming rules of Oracle SQL identifier. Maximum length of vector store name is 125 characters.

This is a mandatory parameter.

attributes

Custom attributes for the vector index in JSON. To see a list of configurable parameters, see Vector Index Attributes.

The default value is NULL.
status
Status of the vector index. The possible values are:
  • Enabled
  • Disabled

The default value is Disabled.

description

Description for the vector index.

The default value is NULL.

Example

The following example demonstrates how to create a vector index and configure the attributes as JSON parameters.

BEGIN                                                                
       DBMS_CLOUD_AI.CREATE_VECTOR_INDEX(                                 
            index_name    => 'MY_INDEX'                                   
            attributes    => JSON_OBJECT(                                 
                       'vector_db_provider' value 'oracle',
                       'vector_table_name'  value 'oracle_mycollection',              
                       'profile_name'      value 'OCIGENAI',         
                       'location'          value                          
                         'https://objectstorage.us-phoenix-1.' ||         
                         'oraclecloud.com/n/mynamespace/b/mybucket',      
                       'object_store_credential_name'   value 'OS_CRED',              
                       'chunk_size'        value 2048,                    
                       'chunk_overlap'     value 256,                     
                       'refresh_rate'      value 720)                     
       );                                                                 
END;                                                                 
/

DROP_VECTOR_INDEX Procedure

This procedure removes a vector store index. It normally removes the vector store index object and deletes the vector store. If set to FALSE, the argument include_data ensures the procedure only removes the vector store index object while retaining the vector store.

Syntax

PROCEDURE DROP_VECTOR_INDEX(                                                
   index_name          IN  VARCHAR2,           
   include_data        IN  BOOLEAN DEFAULT TRUE,                            
   force               IN  BOOLEAN DEFAULT FALSE                           
);

Parameters

Parameter Description

index_name

Name of the vector index. The vector index name must follow the naming rules of Oracle SQL identifier. Maximum length of vector store name is 125 characters.

This is a mandatory parameter.

include_data

Indicates whether to delete both the customer's vector store and vector index along with the vector index object.

Possible values:

  • TRUE
  • FALSE

The default value is TRUE.

force

Indicates whether to ignore errors that occur if the vector index does not exist.

Possible values:

  • TRUE
  • FALSE

If set to TRUE, this parameter bypasses any errors that occur if the vector index does not exist.

The default value is FALSE.

Example

BEGIN
DBMS_CLOUD_AI.DROP_VECTOR_INDEX(
        index_name     => 'MY_INDEX',
        include_data   => FALSE,
        force          => TRUE
     );                                                                 
END;                                                                 
/

DISABLE_VECTOR_INDEX Procedure

This procedure disables a vector index object in the current database. When disabled, an AI profile cannot use the vector index, and the system does not load data into the vector store as new data is added to the object store and does not perform indexing, searching or querying based on the index.

Syntax

DBMS_CLOUD_AI.DISABLE_VECTOR_INDEX(                                              
   index_name       IN  VARCHAR2                                            
);

Parameters

Parameter Description

index_name

Name of the vector index. The vector index name must follow the naming rules of Oracle SQL identifier. Maximum length of vector store name is 125 characters.

This is a mandatory parameter.

Example

You can disable a vector index by providing the name of the vector index.

BEGIN                                                                
   DBMS_CLOUD_AI.DISABLE_VECTOR_INDEX(index_name => 'MY_INDEX');       
END;                                                                 
/

ENABLE_VECTOR_INDEX Procedure

This procedure enables or activates a previously disabled vector index object. Generally, when you create a vector index, by default it is enabled such that the AI profile can use it to perform indexing and searching.

When enabled, a vector index allows an AI profile to use it for loading new data from an object store into a vector store at a user-specified refresh rate. You can specify the refresh_rate parameter through the JSON object list. To configure the JSON attributes, see Vector Index Attributes.

Syntax

DBMS_CLOUD_AI.ENABLE_VECTOR_INDEX(                                              
   index_name       IN  VARCHAR2                                            
);

Parameters

Parameter Description

index_name

Name of the vector index. The vector index name must follow the naming rules of Oracle SQL identifier. Maximum length of vector store name is 125 characters.

This is a mandatory parameter.

Example

You can enable or activate a vector index by specifying the vector index name as follows:

BEGIN                                                                
   DBMS_CLOUD_AI.ENABLE_VECTOR_INDEX(index_name => 'MY_INDEX');       
END;                                                                 
/

UPDATE_VECTOR_INDEX Procedure

This procedure updates an existing vector store index with a specified value of the vector index attribute.
It is overloaded to accept:
  • attribute values of various types.
  • vector index attributes as a JSON document and updates one or more attributes of an existing vector store index with the specified attribute name and value pair.

Syntax

DBMS_CLOUD_AI.UPDATE_VECTOR_INDEX(
   index_name         IN  VARCHAR2,
   attributes         IN  CLOB
);

DBMS_CLOUD_AI.UPDATE_VECTOR_INDEX(
     index_name         IN  VARCHAR2,
     attribute_name     IN  VARCHAR2,
     attribute_value    IN  VARCHAR2
);

DBMS_CLOUD_AI.UPDATE_VECTOR_INDEX(                                                 
     index_name         IN  VARCHAR2,                                         
     attribute_name     IN  VARCHAR2,                                         
     attribute_value    IN  CLOB     DEFAULT NULL                           
  );

Parameters

Parameter Description

index_name

Name of the vector index. The vector index name must follow the naming rules of Oracle SQL identifier. Maximum length of vector store name is 125 characters.

This is a mandatory parameter.

attributes

Specifies vector index attributes in JSON format.

This is a mandatory parameter.

attribute_name

Name of the custom attributes specified as JSON parameters in CREATE_VECTOR_INDEX procedure.

You cannot modify the following attributes:

  • location
  • vector_table_name
  • chunk_size
  • chunk_overlap
  • vector_distance_metric
  • vector_dimension

This is a mandatory parameter.
attribute_value

User specified value for the custom attribute_name. Value can be CLOB, NUMBER or VARCHAR2 type.

The default value is NULL.

Note:

Use either the attributes parameter to specify attribute_name and value pairs in JSON format or the attribute_name and attribute_value parameters together.

Examples

BEGIN                                                                
   DBMS_CLOUD_AI.UPDATE_VECTOR_INDEX(                                 
      index_name       => 'MY_INDEX',                                   
      attribute_name   => 'object_storage_credential_name',                
      attribute_value  => 'NEW_CRED'                           
   );                                                                 
END;                                                                 
/

The following example accepts NUMBER type as the attribute_value. 

BEGIN                                                                
   DBMS_CLOUD_AI.UPDATE_VECTOR_INDEX(                                 
      index_name       => 'MY_INDEX',                                   
      attribute_name   => 'match_limit',                
      attribute_value  => 10                           
   );                                                                 
END;                                                                 
/

The following example accepts VARCHAR2 type as the attribute_value. 

BEGIN                                                                
   DBMS_CLOUD_AI.UPDATE_VECTOR_INDEX(                                 
      index_name       => 'MY_INDEX',                                   
      attribute_name   => 'profile_name',                
      attribute_value  => 'AI_PROF2'                           
   );                                                                 
END;                                                                 
/

The following example accepts attributes in JSON format. 

BEGIN
  DBMS_CLOUD_AI.UPDATE_VECTOR_INDEX(
    index_name => 'MY_VECTOR_INDEX', 
    attributes => '{"match_limit": 10, 
                    "refresh_rate": 30}'
  );
END;
/

Vector Index Attributes

Attributes of a vector index help to manage and configure the behavior of the vector index. You can add custom index attributes as necessary. Some attributes are optional and have a default value.

Attributes

Attribute Name Value Mandatory Description
chunk_size 1024 (default) No

Text size of chunking the input data.

For text data, this means the number of characters.
chunk_overlap 128 (default) No

Specifies the amount of overlapping characters between adjacent chunks of text. This attribute is useful for ensuring contextual continuity and accuracy in text processing by allowing overlaps between segments, which helps prevent loss of contextual information at chunk boundaries.

location

 NA Yes

This parameter specifies source file URI or directories and source files.

Wildcard patterns are supported for both source file URIs and directories.

Cloud source file URIs:

You can specify a source file URI for bucket or subfolder. You can use wildcards to specify subfolders or file names. The character "*" can be used as the wildcard for multiple characters, and the character "?" can be used as the wildcard for a single character.

Example using wild cards:

location_uri => 'https://objectstorage.my$region.oraclecloud.com/n/namespace-string/b/bucketname/o/year=????/month=??/*.csv

The format of the URIs depends on the Cloud Object Storage service you are using, for details see Cloud Object Storage URI Formats.

Directory:

You can specify one directory and file name. The format to specify a directory is: MY_DIR:filename.ext. By default the directory name MY_DIR is a database object and is case-insensitive. The file name is case sensitive.

You can only use wildcards to specify file names in a directory. The character * can be used as the wildcard for multiple characters, and the character ? can be used as the wildcard for a single character. For example: MY_DIR:* or MY_DIR:test?.

Use double quotes to specify a case-sensitive directory name. For example: "my_dir1":*, "my_dir2":Test?

To include a quote character, use two quotes. For example: MY_DIR:''filename.ext. This specifies the filename starts with a quote (').

The files in this location can be documents in formats such as PDF, DOC, JSON, XML, or HTML. See Supported Document Formats.

match_limit

 5 (default) No

Specifies the maximum number of results to return in a vector search query, controlling the output size and improving the efficiency of data retrieval operations.

object_storage_credential_name

 NA Yes

Specifies the name of the credentials for accessing an object storage.

pipeline_name

 <vector_index_name>$VECPIPELINE No

Specifies the name of the vector index data load pipeline. This attribute is automatically set for the vector index, you cannot specify or modify. The pipeline name can be used to monitor the vector index data load using Monitor and Troubleshoot Pipelines.

profile_name

 NA Yes

Name of the AI profile which is used for embedding source data and user prompts.

refresh_rate

 1440 minutes (default) No

Interval of updating data in the vector store. The unit is minutes.

similarity_threshold

 0 (default) No

Defines the minimum level of similarity required for two items to be considered a match, useful for filtering results in matching algorithms to ensure relevance.

vector_distance_metric

A string corresponding to one of the values specified in the description.

No

Specifies the type of distance calculation used to compare vectors in a database, determining how similarity between items is quantified.

Valid values for Oracle 23ai:

  • EUCLIDEAN
  • L2_SQUARED (EUCLIDEAN_SQUARED)
  • COSINE (default)
  • DOT
  • MANHATTAN
  • HAMMING

vector_db_provider

oracle

 Yes

Specifies the provider name that manages and serves as the vector store.

vector_dimension

 NA No

Specifies the number of elements in each vector within the vector store, defining the size and structure of the data representation.

vector_table_name

<vector_index_name>$VECTAB (default)

No

Specifies the name of the table or collection to store vector embeddings and chunked data.
Example: Specify Object Storage URI Location
The following example demonstrates creating a vector index with OCI Generative AI vector store. 
BEGIN
       DBMS_CLOUD_AI.CREATE_VECTOR_INDEX(
         index_name  => 'MY_INDEX',
         attributes  => '{"vector_db_provider": "oracle",
                          "location": "https://swiftobjectstorage.us-phoenix-1.oraclecloud.com/v1/my_namespace/my_bucket/my_data_folder",
                          "object_storage_credential_name": "OCI_CRED",
                          "profile_name": "OPENAI_ORACLE",
                          "vector_dimension": 1024,
                          "vector_distance_metric": "cosine",
                          "chunk_overlap":128,
                          "chunk_size":1024
      }');
END;
/                                                                 
/
Example:Specify Object Storage URI Location with Wild Card Pattern

This example specifies a wild card pattern (*) in the Object Storage URI as the location parameter. It loads all the CSV files from the Object Storage URI. 

BEGIN                                                               
       DBMS_CLOUD_AI.CREATE_VECTOR_INDEX(                                
            index_name    => 'MY_INDEX',                                  
            attributes    => JSON_OBJECT(                                
                       'vector_db_provider' value 'oracle',
                       'vector_table_name'  value 'oracle_mycollection',             
                       'profile_name'      value 'OCIGENAI',        
                       'location'          value 'https://objectstorage.myregion.oraclecloud.com/n/my$namespace/b/bucketname/o/year=????/month=??/file*.csv)',
                       'object_storage_credential_name'   value 'OS_CRED',             
                       'chunk_size'        value 2048,                   
                       'chunk_overlap'     value 256,                    
                       'refresh_rate'      value 720)                    
       );                                                                
END;                                                                
/
Example: Specify Directory Object Location with Wild Card Pattern

This example specifies directory objects in the location parameter using a wild card pattern. It loads all CSV files in the MY_DIR directory. 

BEGIN                                                               
       DBMS_CLOUD_AI.CREATE_VECTOR_INDEX(                                
            index_name    => 'MY_INDEX',                                  
            attributes    => JSON_OBJECT(                                
                       'vector_db_provider' value 'oracle',
                       'vector_table_name'  value 'oracle_mycollection',             
                       'profile_name'      value 'OCIGENAI',        
                       'location'          value 'MY_DIR:*.csv',
                       'object_storage_credential_name'   value 'OS_CRED',             
                       'chunk_size'        value 2048,                   
                       'chunk_overlap'     value 256,                    
                       'refresh_rate'      value 720)                    
       );                                                                
END;                                                                
/
Example: Specify Case-Sensitive Directory Object Location with Wild Card Pattern

This example specifies a case-sensitive directory objects in the location parameter using a wild card pattern. It loads all CSV files in the My_Dir directory. 

BEGIN                                                               
       DBMS_CLOUD_AI.CREATE_VECTOR_INDEX(                                
            index_name    => 'MY_INDEX',                                  
            attributes    => JSON_OBJECT(                                
                       'vector_db_provider' value 'oracle',
                       'vector_table_name'  value 'oracle_mycollection',             
                       'profile_name'      value 'OCIGENAI',        
                       'location'          value '"My_Dir":*.csv',
                       'object_storage_credential_name'   value 'OS_CRED',             
                       'chunk_size'        value 2048,                   
                       'chunk_overlap'     value 256,                    
                       'refresh_rate'      value 720)                    
       );                                                                
END;                                                                
/
Example: Specify Case-Sensitive Directory Object with all Files as Wild Card Pattern

This example specifies a case-sensitive directory object in the location parameter using a wildcard pattern (*). It loads all files located in the My_Dir directory. 

BEGIN                                                               
       DBMS_CLOUD_AI.CREATE_VECTOR_INDEX(                                
            index_name    => 'MY_INDEX',                                  
            attributes    => JSON_OBJECT(                                
                       'vector_db_provider' value 'oracle',
                       'vector_table_name'  value 'oracle_mycollection',             
                       'profile_name'      value 'OCIGENAI',        
                       'location'          value '"My_Dir":*',
                       'object_storage_credential_name'   value 'OS_CRED',             
                       'chunk_size'        value 2048,                   
                       'chunk_overlap'     value 256,                    
                       'refresh_rate'      value 720)                    
       );                                                                
END;                                                                
/
Example: Specify a File Name Match in the Directory Object

This example specifies a directory object and uses a file name prefix, such as test, in the location parameter. It loads all files in the MY_DIR directory whose names begin with test. 

BEGIN                                                               
       DBMS_CLOUD_AI.CREATE_VECTOR_INDEX(                                
            index_name    => 'MY_INDEX',                                  
            attributes    => JSON_OBJECT(                                
                       'vector_db_provider' value 'oracle',
                       'vector_table_name'  value 'oracle_mycollection',             
                       'profile_name'      value 'OCIGENAI',        
                       'location'          value 'MY_DIR:test*',
                       'object_storage_credential_name'   value 'OS_CRED',             
                       'chunk_size'        value 2048,                   
                       'chunk_overlap'     value 256,                    
                       'refresh_rate'      value 720)                    
       );                                                                
END;                                                                
/

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;

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;

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:
  1. Value from the attributes argument in DBMS_CLOUD_AI.GENERATE
  2. Value set in the conversation takes the next priority
  3. 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;

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.

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.

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

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
--------------------------------------------------------------------------------

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.

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.

FEEDBACK Procedure

This procedure enables you to provide feedback to Select AI to potentially improve query generation accuracy. You have the option to provide positive or negative feedback, as well as textual comments or revised SQL queries.

Syntax

DBMS_CLOUD_AI.FEEDBACK(
      profile_name      IN  VARCHAR2,
      sql_id            IN  DBMS_ID,
      feedback_type     IN  VARCHAR2 DEFAULT NULL,
      response          IN  CLOB DEFAULT NULL,
      feedback_content  IN  CLOB DEFAULT NULL,   
      operation         IN  VARCHAR2 DEFAULT 'ADD'
  );
 
DBMS_CLOUD_AI.FEEDBACK(
      profile_name      IN  VARCHAR2,
      sql_text          IN  CLOB,
      feedback_type     IN  VARCHAR2 DEFAULT NULL,
      response          IN  CLOB DEFAULT NULL,
      feedback_content  IN  CLOB DEFAULT NULL,
      operation         IN  VARCHAR2 DEFAULT 'ADD'
  );

Parameters

Parameter Description

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.

This is a mandatory parameter.

sql_id

Identifies the SQL query. One sql_id can have only one feedback entry.

This is a mandatory parameter.

sql_text

Contains the full text of the SQL query.

This is a mandatory parameter.

feedback_type
Specifies the type of feedback. The available values are:
  • positive: Accept the generated SQL.
  • negative: Provide the necessary SQL query improvements by identifying errors in the query.

Note:

The DBMS_CLOUD_AI.FEEDBACK procedure enables you to specify sql_id or sql_text, therefore, the feedback_type is necessary whereas if you are using the feedback action, the LLM dynamically determines or interprets the feedback type.

This is a mandatory parameter when operation is add.

response

Represents the correct SQL query result the user expects.

This is a mandatory parameter when operation is add and feedback_type is negative.

feedback_content

Captures the user's natural language feedback. You have the option to use this parameter along with response.

operation
Specifies the operation to perform. The accepted values are:
  • add (default): Add your feedback by specifying the feedback_type.
  • delete: Delete your feedback by providing the sql_id

Example

Example: Provide Feedback for the Generated SQL Using Add or Delete Operations

The following example demonstrates using the DBMS_CLOUD_AI.FEEDBACK procedure to accept or improve the generated SQL by specifying the parameters from the procedure. 

EXEC DBMS_CLOUD_AI.FEEDBACK(profile_name=>'OCI_FEEDBACK1',
                                   sql_id=> '852w8u83gktc1',
                                   feedback_type=>'positive',
                                   operation=>'add');
EXEC DBMS_CLOUD_AI.FEEDBACK(profile_name=>'OCI_FEEDBACK1',
                                   sql_text=> 'select ai showsql how many movies',
                                   feedback_type=> 'negative',
                                   response=>'SELECT SUM(1) FROM "ADB_USER"."MOVIES"',
                                   feedback_content=>'Use SUM instead of COUNT');
EXEC DBMS_CLOUD_AI.FEEDBACK(profile_name=>'OCI_FEEDBACK1',
                                   sql_id=> '852w8u83gktc1',
                                   operation=>'delete');

Vector Index for FEEDBACK

Select AI creates a default vector index named <profile_name>_FEEDBACK_VECINDEX with default attributes when you use the feedback feature for the first time.

You can modify its attributes such as similarity_threshold and match_limit by using the DBMS_CLOUD_AI.UPDATE_VECTOR_INDEX procedure. This index helps refine future generated SQL based on the feedback provided. This table is dropped when the associated AI profile is dropped. You can also drop <profile_name>_FEEDBACK_VECINDEX. When you do, Select AI no longer uses feedback as hints for the runsql, showsql, and explainsql actions. However, if you submit new feedback using the Select AI feedback feature, Select AI automatically creates a new feedback vector index

Note:

The default value of match_limit for feedback is 3.

Vector Table Name

The table <profile_name>_FEEDBACK_VECINDEX$VECTAB contains vector representations (embeddings) of user feedback along with other parameters, which Select AI uses to improve SQL generation over time.

Parameters

Column Description

attributes

Includes JSON object attributes as per the FEEDBACK Procedure.

content

Contains the user prompt.

embedding

Contains vector representations (embeddings) of user prompt.

Example

The following example demonstrates using the automatically generated vector index table to query and provide feedback.

SQL> select content, attributes from OCI_FEEDBACK1_FEEDBACK_VECINDEX$VECTAB where JSON_VALUE(attributes, '$.sql_text') = 'select ai showsql how many movies';
 
CONTENT                                                
----------------------------------------------------------------------------------------------------
how many movies                                             
ATTRIBUTES
----------------------------------------------------------------------------------------------------
 
{"response":"SELECT SUM(1) FROM \"ADB_USER\".\"MOVIES\"","feedback_type":"negative","sql_id":null,"sql_text":"select ai showsql how many movies","feedback_content":null}
DBMS_CLOUD_AI.feedback Procedure(Positive Feedback)