About Indexing a Dataset for Oracle Analytics AI Assistant

Index a dataset to make its data available to Oracle Analytics AI Assistant. You must index a dataset before users can use its data to ask questions and generate visualizations with the Assistant.

The AI Assistant is available in all regions in the OC1 realm. For more information, see About Regions and Availability Domains. If you don't see the Assistant option, you might not have the required permissions. Contact your administrator to request the Use Assistant in Workbooks permission. See Grant or Revoke Permissions for Application Roles.

You can index any dataset that you have Full Control or Read-Write access to. When you index dataset, you specify if you want to index all or some of a dataset's attributes by names, or by names and values, and which language to index the dataset in. You also select whether the indexed data is available for the home page search, within workbooks, or both.

Configure Dataset Indexing

When you index a dataset, you specify:
  • The attributes to index.
  • Whether to index attribute names only, or attribute names and values.
  • The language to use for the dataset index.
  • Whether to make indexed data available to the Assistant in workbooks, from the Home page, or both.

You can index all attributes or select only the attributes that are useful for answering business questions. You can add synonyms to columns to help the Assistant recognize familiar business terms that users might use instead of a column’s actual name. To learn more about using synonyms, see Make Analytics Content Easier to Search.

For example, you might add Employee Count, Employees, Workforce Size, and Person Count as synonyms for a column named Headcount.

For a file-based dataset, you must index the dataset before you or users that you share the dataset with can use it to generate visualizations from a home page.

Use the Search tab on the dataset’s Inspect page to configure dataset indexing. Administrators use the Data Model pane on the Console's Search Index page to configure indexing for subject areas. See Configure Search Indexing

Improve Assistant responses with AI Descriptions

Use AI Descriptions to add semantic metadata that explains the business meaning of a dataset and its columns. Oracle Analytics AI Assistant uses this context when a user asks a semantically related question, helping it interpret the data more accurately and consistently.

Column names and synonyms help the Assistant identify the appropriate column. AI Descriptions provide additional context that explains how the Assistant should interpret a column’s values. For example, an AI Description can:

  • Distinguish between similar columns.
  • Explain a measure’s grain, unit, or scale.
  • Describe how values are formatted.
  • Explain relationships between columns.
  • Identify important inclusions, exclusions, or aggregation rules.

For example, a dataset might contain both Invoice Forecast and Opportunity Forecast columns. Add an AI Description to each column to explain which business process the forecast represents. This helps the Assistant select the correct measure for a user’s question.

You can also use an AI Description to prevent incorrect aggregation. For example, if a Consolidated Total invoice type already includes the other invoice types, describe that rule so the Assistant doesn’t combine the consolidated value with its component values.

AI Descriptions define dataset semantics. They don't replace Agent supplemental instructions, which define Agent behavior and guidance for responding to users. An AI Description belongs to the dataset or column and can be used by multiple Agents.

You can add descriptions manually or select Generate to create suggested descriptions. Review generated descriptions and retain only information that adds meaningful business context beyond column names and synonyms. See About AI Descriptions for Datasets and Add AI Descriptions to a Dataset.

Keep Dataset Indexes Current

For file-based datasets, indexes are refreshed when you reload the data manually or according to a schedule that you specify.

By default, Oracle Analytics indexes a dataset whenever its data is refreshed, so that current data is available when users search from the Home page. Dataset data is automatically refreshed whenever a user saves changes to the dataset.

Alternatively, you can schedule indexing or index a dataset on demand.

Security and Data Privacy

Assistant requests are processed in a secure OCI environment. However, an Assistant request might be routed to another OCI region for processing.

Oracle Analytics uses foundation models deployed through OCI Generative AI Service. These models are used in their native state, without fine-tuning or model customization. Oracle doesn’t use your data to train the models; Oracle Analytics achieves its results through prompt engineering.

Verify Assistant -Generated Results

Oracle Analytics AI Assistant uses a large language model (LLM) to process questions. As a result, it might generate responses based on patterns in the data rather than factual accuracy. Always verify Assistant-generated results and consult your primary data sources before making important business decisions.

Avoid entering personally identifying information in natural-language prompts. Prompt text is sent to the LLM exactly as written.