Oracle Cloud Infrastructure Documentation

Creating a Vector Store

Create a vector store in the OCI Generative AI service.

This task documents the steps for creating a vector store by using the Oracle Cloud Console.

Tip

After you create a vector store, you can view its details and perform other tasks, such as updating it or deleting it. Use the Actions menu (three dots) in the Console to access these tasks.

To create a vector store, in the list page, select Create vector store. If you need help finding the list page, see Listing Vector Stores.

Basic Information

  1. Enter a name for the vector store.
    Start the name with a letter or underscore, followed by letters, numbers, hyphens, or underscores. The length can be 1 to 255 characters.
  2. (Optional) Enter a description for the vector store.
  3. Select a compartment to create the vector store in. The default compartment is the same as the list page, but you can select any compartment that you have permission to work in.

Data Source Type

Select one of the following options:
  • Unstructured data

    Upload documents such as PDFs and text files. Use for semantic search and RAG.

  • Structured data

    Create a semantic store to use for NL2SQL and schema-aware querying.

Unstructured Data

If you selected Unstructured data, follow these steps:

  1. Enable the expiration policy and specify the inactivity period (in days) before the vector store is deleted.

    Default: 100 days.

  2. (Optional) Expand Metadata and add up to 16 metadata key-value pairs. Use metadata to filter data sources and limit agent search to relevant files.

Structured Data

If you selected Structured data, follow these steps.

  1. Under Configure sync connector, select a Connection type.

    Allowed value: OCI Database tool.

  2. Enter the Enrichment connection id, then select Test enrichment connection to verify access.
  3. Enter the Querying connection id, then select Test query connection to verify access.
  4. In Schema, specify the database schema names to ingest for NL2SQL and schema-aware querying.
  5. For Enrichment, select when to enrich the semantic store.
    • None: Create the semantic store without starting enrichment. You can run enrichment later.
    • On create: Start enrichment after the semantic store is created.
    • Interval: Refresh the enriched metadata on a recurring schedule. In Refresh schedule interval detail, enter an ISO 8601 duration. The minimum interval is six hours. For example, enter PT6H to refresh the enriched metadata every six hours or P1D to refresh it once a day. Each refresh uses the selected Generative AI model. For more duration formats and examples, see ISO 8601 durations.
  6. (Optional) Enable Include user-defined semantics to enrich the semantic store with database annotations, comments, and synonyms in addition to metadata.
  7. Under Model configuration, select a Generative AI model type.

    Dedicated AI endpoints and user-selected embedding models aren't supported. The embedding model is managed by the service.

    Note

    OpenAI gpt-oss-120b is the default model and has been benchmarked specifically for NL2SQL. Other supported on-demand models have been assessed for use with OCI Generative AI, but their NL2SQL-specific accuracy and performance haven't been benchmarked. We recommend evaluating the selected model with your schema and workload before using it in production.
    • Default: Use the default NL2SQL model, openai.gpt-oss-120b, for enrichment.
    • Custom: In Generative AI model, select an on-demand model available in the region and accessible to your tenancy. See Generative AI Models by Region.

Tags

  1. (Optional) Select Add tag and assign tags to this vector store. See Resource Tags.
  2. Select Create.