1. Vector Index¶

VectorIndex supports Retrieval Augmented Generation (RAG). It converts source documents into vector embeddings, stores the embeddings in a vector store, and links the vector index to a Select AI profile. When that profile is used for natural language generation, Select AI can retrieve semantically similar content from the vector index and use that content as grounding context for the response.

A vector index is useful when the answer should come from files or documents that are not represented as relational tables. Typical sources include documents in Object Storage, product manuals, generated reports, logs, JSON files, and other text-heavy content that should be searched by meaning rather than exact keywords.

Before creating a vector index, make sure the database user has:

A Select AI profile with an AI provider that supports embeddings.
A credential for the AI provider used by the profile.
A credential for the object storage location if the source objects are not public.
Network access to the AI provider endpoint and the source location. See Privileges for network ACL helpers.

The usual lifecycle is:

Create a profile with a provider and embedding model.
Create OracleVectorIndexAttributes with the source location and storage credential.
Create VectorIndex and call create().
Use the linked profile for RAG actions such as narrate().
Fetch, list, update, disable, enable, or delete the index as needed.

2. `VectorIndex` Object Model¶

3. `VectorIndexAttributes`¶

A VectorIndexAttributes object can be created with select_ai.VectorIndexAttributes(). Also check vector index attributes

For Oracle vector indexes, use OracleVectorIndexAttributes. It sets vector_db_provider to VectorDBProvider.ORACLE and is the preferred attribute class for the examples in this guide.

Common attributes:

Attribute	Use
`location`	Object Storage URI or source location containing the documents to embed.
`object_storage_credential_name`	Credential used to read the source location.
`profile_name`	Select AI profile used to create embeddings and answer RAG prompts. If omitted during `create()`, it is taken from the `profile` object passed to `VectorIndex`.
`chunk_size` and `chunk_overlap`	Control how source text is split before embedding. Larger chunks keep more context together; overlap helps preserve context across chunk boundaries.
`match_limit`	Maximum number of matching chunks returned during semantic search.
`similarity_threshold`	Minimum similarity score required for retrieved chunks to be considered relevant.
`vector_distance_metric`	Distance metric used to compare embeddings. Supported values include `COSINE`, `EUCLIDEAN`, `L2_SQUARED`, `DOT`, `MANHATTAN`, and `HAMMING`.
`refresh_rate`	Refresh interval, in minutes, for loading new or changed source data.
`vector_table_name`	Name of the table used to store vector embeddings and chunked data. Leave unset unless you need to control the storage table name.
`enable_sources`	Include filenames and source links in RAG output when supported by the profile and model response.

Example attributes:

attributes = select_ai.OracleVectorIndexAttributes(
    location="https://objectstorage.us-ashburn-1.oraclecloud.com/n/example/b/docs/o/product-guides",
    object_storage_credential_name="object_store_credential",
    chunk_size=1024,
    chunk_overlap=128,
    match_limit=5,
    similarity_threshold=0.5,
    vector_distance_metric=select_ai.VectorDistanceMetric.COSINE,
    refresh_rate=1440,
)

The embedding model is configured on the provider inside the linked ProfileAttributes. Keep the profile provider and vector index attributes together conceptually: the profile decides how embeddings are generated, while the vector index attributes decide where content is read from, how it is chunked, and how the vector store is searched.

Attributes of a vector index help to manage and configure the behavior of the vector index.

Parameters:

chunk_size (int) – Text size of chunking the input data.
chunk_overlap (int) – Specifies the amount of overlapping characters between adjacent chunks of text.
enable_sources – Provides document source links and filenames in RAG output
location (str) – Location of the object store.
match_limit (int) – Specifies the maximum number of results to return in a vector search query
object_storage_credential_name (str) – Name of the credentials for accessing object storage.
profile_name (str) – Name of the AI profile which is used for embedding source data and user prompts.
refresh_rate (int) – Interval of updating data in the vector store. The unit is minutes.
similarity_threshold (float) – Defines the minimum level of similarity required for two items to be considered a match
vector_distance_metric (select_ai.vector_index.VectorDistanceMetric) – Specifies the type of distance calculation used to compare vectors in a database
vector_db_provider (select_ai.vector_index.VectorDBProvider) – Name of the Vector database provider. Default value is “oracle”
vector_db_endpoint (str) – Endpoint to access the Vector database
vector_db_credential_name (str) – Name of the credentials for accessing Vector database
vector_dimension (int) – Specifies the number of elements in each vector within the vector store
vector_table_name (str) – Specifies the name of the table or collection to store vector embeddings and chunked data

chunk_overlap: int | None = None¶

chunk_size: int | None = None¶

classmethod create(*, vector_db_provider: str | None = None, **kwargs)¶

dict(exclude_null=True)¶

enable_sources: bool | None = None¶

json(exclude_null=True)¶

classmethod keys()¶

location: str | None = None¶

match_limit: int | None = None¶

object_storage_credential_name: str | None = None¶

pipeline_name: str | None = None¶

profile_name: str | None = None¶

refresh_rate: int | None = None¶

similarity_threshold: float | None = None¶

vector_db_credential_name: str | None = None¶

vector_db_endpoint: str | None = None¶

vector_db_provider: VectorDBProvider | None = None¶

vector_dimension: int | None = None¶

vector_distance_metric: VectorDistanceMetric | None = None¶

vector_table_name: str | None = None¶

3.1. `OracleVectorIndexAttributes`¶

Oracle specific vector index attributes

chunk_overlap: int | None = None¶

chunk_size: int | None = None¶

classmethod create(*, vector_db_provider: str | None = None, **kwargs)¶

dict(exclude_null=True)¶

enable_sources: bool | None = None¶

json(exclude_null=True)¶

classmethod keys()¶

location: str | None = None¶

match_limit: int | None = None¶

object_storage_credential_name: str | None = None¶

pipeline_name: str | None = None¶

profile_name: str | None = None¶

refresh_rate: int | None = None¶

similarity_threshold: float | None = None¶

vector_db_credential_name: str | None = None¶

vector_db_endpoint: str | None = None¶

vector_db_provider: VectorDBProvider | None = 'oracle'¶

vector_dimension: int | None = None¶

vector_distance_metric: VectorDistanceMetric | None = None¶

vector_table_name: str | None = None¶

4. `VectorIndex` API¶

A VectorIndex object can be created with select_ai.VectorIndex()

VectorIndex objects let you manage vector indexes

Parameters:

index_name (str) – The name of the vector index
description (str) – The description of the vector index
attributes (select_ai.VectorIndexAttributes) – The attributes of the vector index

create(replace: bool | None = False, wait_for_completion: bool = False)¶

Create a vector index in the database and populates the index: with data from an object store bucket using an async scheduler job

Parameters:

replace (bool) – Replace vector index if it exists
wait_for_completion (bool) – True to wait for index creation

Returns:

None

delete(include_data: bool | None = True, force: bool | None = False)¶

This procedure removes a vector store index

Parameters:

include_data (bool) – Indicates whether to delete both the customer’s vector store and vector index along with the vector index object
force (bool) – Indicates whether to ignore errors that occur if the vector index does not exist

Returns:

None

Raises:

oracledb.DatabaseError

classmethod delete_index(index_name: str, include_data: bool = True, force: bool = False)¶

Class method to remove a vector store index

Parameters:

index_name (str) – The name of the vector index
include_data (bool) – Indicates whether to delete both the customer’s vector store and vector index along with the vector index object
force (bool) – Indicates whether to ignore errors that occur if the vector index does not exist

Returns:

None

Raises:

oracledb.DatabaseError

disable()¶

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.

Returns:: None
Raises:: oracledb.DatabaseError

enable()¶

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.

Returns:: None
Raises:: oracledb.DatabaseError

classmethod fetch(index_name: str, owner: str | None = None) → VectorIndex¶

Fetches vector index attributes from the database and builds a proxy object for the passed index_name

Parameters:

index_name (str) – The name of the vector index
owner (str) – Owner of a shared index. Defaults to current schema.

get_attributes() → VectorIndexAttributes¶

Get attributes of this vector index

Returns:: select_ai.VectorIndexAttributes
Raises:: VectorIndexNotFoundError

get_next_refresh_timestamp() → datetime | None¶: Returns the UTC timestamp of the next scheduled refresh

get_profile() → Profile¶

Get Profile object linked to this vector index

Returns:: select_ai.Profile
Raises:: ProfileNotFoundError

grant_access(user_or_role_name: str) → None¶: Grant a database user or role access to this vector index.

classmethod list(index_name_pattern: str = '.*', owner: str | None = None) → Iterator[VectorIndex]¶

List Vector Indexes

Parameters:

index_name_pattern (str) – Regular expressions can be used to specify a pattern. Function REGEXP_LIKE is used to perform the match. Default value is “.*” i.e. match all vector indexes.
owner (str) – Owner of shared indexes. Defaults to current schema.

Returns:

Iterator[VectorIndex]

property qualified_name: str | None¶: Return the owner-qualified index name when owner is known.

revoke_access(user_or_role_name: str) → None¶: Revoke a database user or role’s vector index access.

set_attribute(attribute_name: str, attribute_value: str | int | float)¶

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

Parameters:

attribute_name (str) – Custom attribute name
attribute_value (Union[str, int, float]) – Attribute Value

set_attributes(attributes: VectorIndexAttributes = None)¶

This procedure updates an existing vector store index with a specified value of the vector index attributes. Specify multiple attributes by passing an object of type :class VectorIndexAttributes

Parameters:: attributes (select_ai.VectorIndexAttributes) – Use this to update multiple attribute values
Returns:: None
Raises:: oracledb.DatabaseError

Use the synchronous API in scripts, notebooks, and command-line tools that use select_ai.connect(). Use AsyncVectorIndex in applications already using asyncio and select_ai.async_connect() or an async connection pool.

Important lifecycle methods:

Method	Use
`create(replace=False, wait_for_completion=False)`	Create the database vector index and start the load pipeline. If `replace=True` and the index already exists, the existing index is dropped and recreated. Use `wait_for_completion=True` when the next step depends on the initial load being complete.
`fetch(index_name, owner=None)`	Build a `VectorIndex` proxy from database metadata, including attributes and the linked profile when it still exists.
`list(index_name_pattern=".*", owner=None)`	Iterate over vector indexes visible to the current user or owned by the specified schema. The pattern is evaluated with Oracle `REGEXP_LIKE`.
`set_attribute()` and `set_attributes()`	Update one or more index attributes.
`get_next_refresh_timestamp()`	Return the next scheduled refresh timestamp in UTC when the index has a refresh rate and a recorded pipeline execution.
`disable()` and `enable()`	Pause or resume use of the vector index for loading, indexing, searching, and querying.
`delete(include_data=True, force=False)`	Drop the vector index. `include_data=True` also removes associated vector store data. `force=True` ignores missing-index errors.

Check the examples below to understand how to create vector indexes.

4.1. Create vector index¶

In the following example, vector database provider is Oracle and objects used to create embeddings reside in OCI Object Storage. The profile uses an OCI Generative AI provider with an embedding model, and the vector index is linked to that profile during create().

Upload samples/data/refund-policy.txt from the repository to a bucket you can read. The document says customers have 30 calendar days to request a refund. Configure the sample with the URI of that uploaded file and the name of an existing database credential authorized to read it:

export SELECT_AI_DOCUMENT_URI='https://objectstorage.REGION.oraclecloud.com/n/NAMESPACE/b/BUCKET/o/refund-policy.txt'
export SELECT_AI_OBJECT_STORAGE_CREDENTIAL='OBJECT_STORE_CRED'
export SELECT_AI_PROVIDER_CREDENTIAL='my_oci_ai_profile_key'

Replace REGION, NAMESPACE, and BUCKET with your storage values. Use the common database connection variables from Connection. The provider credential and storage credential can be the same credential when it has both permissions. Optionally set SELECT_AI_OCI_REGION and SELECT_AI_EMBEDDING_MODEL for your OCI configuration. The default region is us-chicago-1 and embedding model is cohere.embed-v4.0; choose a model available in your account and region.

The creation samples replace test_vector_index and their named RAG profile. Use a development schema, or edit these names before running if you need to preserve existing definitions. The samples close their database connections but leave the profile and index available for the next step.

From the repository root, run python samples/vector_index_create.py. For async, use python samples/async/vector_index_create.py. Both examples wait for the initial load to complete. Run either workflow in its own schema, or delete the existing sample index before switching between them.

import os

import select_ai

user = os.environ["SELECT_AI_USER"]
password = os.environ["SELECT_AI_PASSWORD"]
dsn = os.environ["SELECT_AI_DB_CONNECT_STRING"]
# Upload samples/data/refund-policy.txt to your bucket, then set its URI.
source_uri = os.environ["SELECT_AI_DOCUMENT_URI"]
provider_credential = os.getenv(
    "SELECT_AI_PROVIDER_CREDENTIAL", "my_oci_ai_profile_key"
)
storage_credential = os.environ["SELECT_AI_OBJECT_STORAGE_CREDENTIAL"]

select_ai.connect(user=user, password=password, dsn=dsn)
try:
    provider = select_ai.OCIGenAIProvider(
        region=os.getenv("SELECT_AI_OCI_REGION", "us-chicago-1"),
        oci_apiformat="GENERIC",
        embedding_model=os.getenv(
            "SELECT_AI_EMBEDDING_MODEL", "cohere.embed-v4.0"
        ),
    )
    profile = select_ai.Profile(
        profile_name="oci_vector_ai_profile",
        attributes=select_ai.ProfileAttributes(
            credential_name=provider_credential,
            provider=provider,
        ),
        description="Profile for the sample refund policy",
        replace=True,
    )
    vector_index = select_ai.VectorIndex(
        index_name="test_vector_index",
        attributes=select_ai.OracleVectorIndexAttributes(
            location=source_uri,
            object_storage_credential_name=storage_credential,
        ),
        description="Sample refund policy",
        profile=profile,
    )
    vector_index.create(replace=True, wait_for_completion=True)
    print("Created vector index: test_vector_index")
finally:
    select_ai.disconnect()

output:

Created vector index: test_vector_index

4.2. List vector index¶

import os

import select_ai

user = os.getenv("SELECT_AI_USER")
password = os.getenv("SELECT_AI_PASSWORD")
dsn = os.getenv("SELECT_AI_DB_CONNECT_STRING")

select_ai.connect(user=user, password=password, dsn=dsn)
vector_index = select_ai.VectorIndex()
for index in vector_index.list(index_name_pattern="^test"):
    print("Vector index", index.index_name)
    print("Vector index profile", index.profile)

The script prints each matching index name and its linked profile. Results depend on the indexes visible to the connected user.

4.3. Fetch vector index¶

You can fetch the vector index attributes and associated AI profile using the class method VectorIndex.fetch(index_name, owner=None). Pass owner for an index shared from another schema. The returned object exposes the resolved owner and the owner-qualified qualified_name. Fetch is useful when the index was created earlier or by another process and you want to inspect or update it without recreating the original Python object.

vector_index = select_ai.VectorIndex.fetch(
    "PRODUCT_DOCS",
    owner="APP_OWNER",
)
print(vector_index.qualified_name)

import os

import select_ai

user = os.getenv("SELECT_AI_USER")
password = os.getenv("SELECT_AI_PASSWORD")
dsn = os.getenv("SELECT_AI_DB_CONNECT_STRING")

select_ai.connect(user=user, password=password, dsn=dsn)

vector_index = select_ai.VectorIndex.fetch(index_name="test_vector_index")
print(vector_index.attributes)
print(vector_index.profile)
print(vector_index.get_next_refresh_timestamp())

The script prints the saved index attributes, its linked profile, and the next refresh timestamp. The URI and credentials reflect your configuration; the next refresh can be None when no refresh execution has been recorded.

4.4. Update vector index attributes¶

To update attributes, use either vector_index.set_attribute() or vector_index.set_attributes(). Use set_attribute() for a single value and set_attributes() when updating several values together.

import os

import select_ai

user = os.getenv("SELECT_AI_USER")
password = os.getenv("SELECT_AI_PASSWORD")
dsn = os.getenv("SELECT_AI_DB_CONNECT_STRING")
select_ai.connect(user=user, password=password, dsn=dsn)
vector_index = select_ai.VectorIndex(
    index_name="test_vector_index",
)

# Use vector_index.set_attributes to update a multiple attributes
updated_attributes = select_ai.OracleVectorIndexAttributes(refresh_rate=1450)
vector_index.set_attributes(attributes=updated_attributes)

# Use vector_index.set_attribute to update a single attribute
vector_index.set_attribute(
    attribute_name="similarity_threshold", attribute_value=0.5
)
print(vector_index.attributes)

The script sets refresh_rate to 1450 minutes and similarity_threshold to 0.5, then prints the attributes held by the Python object.

4.5. RAG using vector index¶

After create() succeeds, the profile has its vector_index_name set to the new index. Use that profile with text-returning actions such as narrate() to retrieve relevant chunks from the vector index and ground the answer in the indexed content.

import os

import select_ai

user = os.environ["SELECT_AI_USER"]
password = os.environ["SELECT_AI_PASSWORD"]
dsn = os.environ["SELECT_AI_DB_CONNECT_STRING"]

select_ai.connect(user=user, password=password, dsn=dsn)
try:
    profile = select_ai.Profile(
        profile_name="oci_vector_ai_profile"
    )
    response = profile.narrate(
        "How many days do customers have to request a refund?"
    )
    print(response)
finally:
    select_ai.disconnect()

The answer should identify the 30 calendar day refund window. Exact wording and source formatting depend on the provider and profile settings. For example:

output:

Customers can request a refund within 30 calendar days of purchase.

If no answer is grounded in the policy, verify that the uploaded URI and storage credential are correct, that the initial load completed, and that this profile is linked to test_vector_index. Run the corresponding RAG script after creating the index. See Delete vector index to remove the sample index and its data; delete the sample profile separately when finished.

4.6. Delete vector index¶

Use delete() when the index is no longer needed. By default, include_data=True removes the vector index metadata and the associated vector store data. Set include_data=False only when you intentionally want to keep the underlying vector store data.

import os

import select_ai

user = os.getenv("SELECT_AI_USER")
password = os.getenv("SELECT_AI_PASSWORD")
dsn = os.getenv("SELECT_AI_DB_CONNECT_STRING")

select_ai.connect(user=user, password=password, dsn=dsn)
vector_index = select_ai.VectorIndex(index_name="test_vector_index")
vector_index.delete(force=True)
print("Deleted vector index: test_vector_index")

output:

Deleted vector index: test_vector_index

5. `AsyncVectorIndex` API¶

An AsyncVectorIndex object can be created with select_ai.AsyncVectorIndex()

AsyncVectorIndex objects let you manage vector indexes using async APIs. Use this for non-blocking concurrent requests

Parameters:

index_name (str) – The name of the vector index
description (str) – The description of the vector index
attributes (VectorIndexAttributes) – The attributes of the vector index

async create(replace: bool | None = False, wait_for_completion: bool | None = False) → None¶

Create a vector index in the database and populates it with data from an object store bucket using an async scheduler job

Parameters:

replace (bool) – True to replace existing vector index
wait_for_completion (bool) – True to wait for index creation

async delete(include_data: bool | None = True, force: bool | None = False) → None¶

This procedure removes a vector store index.

Parameters:

include_data (bool) – Indicates whether to delete both the customer’s vector store and vector index along with the vector index object.
force (bool) – Indicates whether to ignore errors that occur if the vector index does not exist.

Returns:

None

Raises:

oracledb.DatabaseError

async classmethod delete_index(index_name: str, include_data: bool = True, force: bool = False)¶

Class method to remove a vector store index

Parameters:

index_name (str) – The name of the vector index
include_data (bool) – Indicates whether to delete both the customer’s vector store and vector index along with the vector index object
force (bool) – Indicates whether to ignore errors that occur if the vector index does not exist

Returns:

None

Raises:

oracledb.DatabaseError

async disable() → None¶

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.

Returns:: None
Raises:: oracledb.DatabaseError

async enable() → None¶

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.

Returns:: None
Raises:: oracledb.DatabaseError

async classmethod fetch(index_name: str, owner: str | None = None) → AsyncVectorIndex¶

Fetches vector index attributes from the database and builds a proxy object for the passed index_name

Parameters:

index_name (str) – The name of the vector index
owner (str) – Owner of a shared index. Defaults to current schema.

async get_attributes() → VectorIndexAttributes¶

Get attributes of a vector index

Returns:: select_ai.VectorIndexAttributes
Raises:: VectorIndexNotFoundError

async get_next_refresh_timestamp() → datetime | None¶: Return the UTC timestamp for the next scheduled refresh.

async get_profile() → AsyncProfile¶

Get AsyncProfile object linked to this vector index

Returns:: select_ai.AsyncProfile
Raises:: ProfileNotFoundError

async grant_access(user_or_role_name: str) → None¶: Asynchronously grant a user or role access to this vector index.

classmethod list(index_name_pattern: str = '.*', owner: str | None = None) → AsyncGenerator[AsyncVectorIndex, None]¶

List Vector Indexes.

Parameters:

index_name_pattern (str) – Regular expressions can be used to specify a pattern. Function REGEXP_LIKE is used to perform the match. Default value is “.*” i.e. match all vector indexes.
owner (str) – Owner of shared indexes. Defaults to current schema.

Returns:

AsyncGenerator[VectorIndex]

property qualified_name: str | None¶: Return the owner-qualified index name when owner is known.

async revoke_access(user_or_role_name: str) → None¶: Asynchronously revoke a user or role’s vector index access.

async set_attribute(attribute_name: str, attribute_value: str | int | float) → None¶

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

Parameters:

attribute_name (str) – Custom attribute name
attribute_value (Union[str, int, float]) – Attribute Value

async set_attributes(attributes: VectorIndexAttributes) → None¶

This procedure updates an existing vector store index with a specified value of the vector index attribute. multiple attributes by passing an object of type :class VectorIndexAttributes

Parameters:: attributes (select_ai.VectorIndexAttributes) – Use this to update multiple attribute values
Returns:: None
Raises:: oracledb.DatabaseError

The async API mirrors the synchronous API. Async profile construction and vector index methods that access the database must be awaited, and AsyncVectorIndex.list() is an async iterator.

Synchronous API	Async API
`select_ai.connect(...)`	`await select_ai.async_connect(...)`
`Profile(...)`	`await AsyncProfile(...)`
`VectorIndex.create(...)`	`await AsyncVectorIndex.create(...)`
`VectorIndex.fetch(...)`	`await AsyncVectorIndex.fetch(...)`
`for index in VectorIndex.list(...)`	`async for index in AsyncVectorIndex.list(...)`
`profile.narrate(...)`	`await async_profile.narrate(...)`

5.1. Async create vector index¶

import asyncio
import os

import select_ai

user = os.environ["SELECT_AI_USER"]
password = os.environ["SELECT_AI_PASSWORD"]
dsn = os.environ["SELECT_AI_DB_CONNECT_STRING"]
# Upload samples/data/refund-policy.txt to your bucket, then set its URI.
source_uri = os.environ["SELECT_AI_DOCUMENT_URI"]
provider_credential = os.getenv(
    "SELECT_AI_PROVIDER_CREDENTIAL", "my_oci_ai_profile_key"
)
storage_credential = os.environ["SELECT_AI_OBJECT_STORAGE_CREDENTIAL"]


async def main():
    await select_ai.async_connect(user=user, password=password, dsn=dsn)
    try:
        provider = select_ai.OCIGenAIProvider(
            region=os.getenv("SELECT_AI_OCI_REGION", "us-chicago-1"),
            oci_apiformat="GENERIC",
            embedding_model=os.getenv(
                "SELECT_AI_EMBEDDING_MODEL", "cohere.embed-v4.0"
            ),
        )
        profile = await select_ai.AsyncProfile(
            profile_name="async_oci_vector_ai_profile",
            attributes=select_ai.ProfileAttributes(
                credential_name=provider_credential,
                provider=provider,
            ),
            description="Profile for the sample refund policy",
            replace=True,
        )
        vector_index = select_ai.AsyncVectorIndex(
            index_name="test_vector_index",
            attributes=select_ai.OracleVectorIndexAttributes(
                location=source_uri,
                object_storage_credential_name=storage_credential,
            ),
            description="Sample refund policy",
            profile=profile,
        )
        await vector_index.create(replace=True, wait_for_completion=True)
        print("Created vector index: test_vector_index")
    finally:
        await select_ai.async_disconnect()


asyncio.run(main())

output:

Created vector index: test_vector_index

5.2. Async list vector index¶

import asyncio
import os

import select_ai

user = os.getenv("SELECT_AI_USER")
password = os.getenv("SELECT_AI_PASSWORD")
dsn = os.getenv("SELECT_AI_DB_CONNECT_STRING")


async def main():
    await select_ai.async_connect(user=user, password=password, dsn=dsn)
    vector_index = select_ai.AsyncVectorIndex()
    async for index in vector_index.list(index_name_pattern="^test"):
        print("Vector index", index.index_name)
        print("Vector index profile", index.profile)


asyncio.run(main())

The script prints matching index names and their linked profiles. Results depend on your schema and configuration.

5.3. Async fetch vector index¶

You can fetch the vector index attributes and associated AI profile using the class method AsyncVectorIndex.fetch(index_name)

import asyncio
import os

import select_ai

user = os.getenv("SELECT_AI_USER")
password = os.getenv("SELECT_AI_PASSWORD")
dsn = os.getenv("SELECT_AI_DB_CONNECT_STRING")


async def main():
    await select_ai.async_connect(user=user, password=password, dsn=dsn)
    async_vector_index = await select_ai.AsyncVectorIndex.fetch(
        index_name="test_vector_index"
    )
    print(async_vector_index.attributes)
    print(async_vector_index.profile)
    print(await async_vector_index.get_next_refresh_timestamp())


asyncio.run(main())

The script prints the saved attributes and linked profile. Source URIs, credentials, and other values depend on the index you created.

5.4. Async update vector index attributes¶

To update attributes, use either async_vector_index.set_attribute() or async_vector_index.set_attributes()

import asyncio
import os

import select_ai

user = os.getenv("SELECT_AI_USER")
password = os.getenv("SELECT_AI_PASSWORD")
dsn = os.getenv("SELECT_AI_DB_CONNECT_STRING")


async def main():
    await select_ai.async_connect(user=user, password=password, dsn=dsn)
    async_vector_index = select_ai.AsyncVectorIndex(
        index_name="test_vector_index",
    )

    # Use vector_index.set_attributes to update a multiple attributes
    updated_attributes = select_ai.OracleVectorIndexAttributes(
        refresh_rate=1450
    )
    await async_vector_index.set_attributes(attributes=updated_attributes)

    # Use vector_index.set_attribute to update a single attribute
    await async_vector_index.set_attribute(
        attribute_name="similarity_threshold", attribute_value=0.5
    )
    print(async_vector_index.attributes)


asyncio.run(main())

The script updates the refresh rate and similarity threshold, then prints the attributes held by the Python object.

5.5. Async RAG using vector index¶

import os

import select_ai

user = os.environ["SELECT_AI_USER"]
password = os.environ["SELECT_AI_PASSWORD"]
dsn = os.environ["SELECT_AI_DB_CONNECT_STRING"]


async def main():
    await select_ai.async_connect(user=user, password=password, dsn=dsn)
    try:
        profile = await select_ai.AsyncProfile(
            profile_name="async_oci_vector_ai_profile"
        )
        response = await profile.narrate(
            "How many days do customers have to request a refund?"
        )
        print(response)
    finally:
        await select_ai.async_disconnect()


asyncio.run(main())

The answer should identify the 30 calendar day refund window. Exact wording and source formatting depend on the provider and profile settings. For example:

output:

Customers can request a refund within 30 calendar days of purchase.

If no answer is grounded in the policy, verify that the uploaded URI and storage credential are correct, that the initial load completed, and that this profile is linked to test_vector_index. Run the corresponding RAG script after creating the index. See Delete vector index to remove the sample index and its data; delete the sample profile separately when finished.

1. Vector Index¶

2. `VectorIndex` Object Model¶

3. `VectorIndexAttributes`¶

3.1. `OracleVectorIndexAttributes`¶

4. `VectorIndex` API¶

4.1. Create vector index¶

4.2. List vector index¶

4.3. Fetch vector index¶

4.4. Update vector index attributes¶

4.5. RAG using vector index¶

4.6. Delete vector index¶

5. `AsyncVectorIndex` API¶

5.1. Async create vector index¶

5.2. Async list vector index¶

5.3. Async fetch vector index¶

5.4. Async update vector index attributes¶

5.5. Async RAG using vector index¶

Table of Contents

Previous topic

Next topic

This Page

1. Vector Index¶

2. VectorIndex Object Model¶

3. VectorIndexAttributes¶

3.1. OracleVectorIndexAttributes¶

4. VectorIndex API¶

4.1. Create vector index¶

4.2. List vector index¶

4.3. Fetch vector index¶

4.4. Update vector index attributes¶

4.5. RAG using vector index¶

4.6. Delete vector index¶

5. AsyncVectorIndex API¶

5.1. Async create vector index¶

5.2. Async list vector index¶

5.3. Async fetch vector index¶

5.4. Async update vector index attributes¶

5.5. Async RAG using vector index¶

5.5.1. Share a vector index¶

2. `VectorIndex` Object Model¶

3. `VectorIndexAttributes`¶

3.1. `OracleVectorIndexAttributes`¶

4. `VectorIndex` API¶

5. `AsyncVectorIndex` API¶